View
31
Download
0
Category
Preview:
Citation preview
1
Switch reference: an overview
Until not very long ago, switch reference was regarded as a marginal phenomenon found in a handful
of lesser-known languages. An increasing number of studies of the phenomenon made it clear,
however, that the geographical extent of switch-reference systems is rather large, spanning large parts
of the New World, Siberia, Oceania, Australia, and also including some areas in Africa and East Asia.
The growing body of information on the topic raises new questions about the development, functions,
and nature of switch reference, as well as the internal variation between different switch reference
systems. This contribution will give an overview of the relevant issues that are involved in switch
reference, focusing on geographical, typological-theoretical, and diachronic aspects, and suggesting a
number of lines for further research.
1. Introduction
The term switch reference (SR) will soon celebrate its 50th anniversary since its coinage in Jacobsen
(1967).1 In half a century of research, SR has evolved from an exotic phenomenon in a handful of
North-American languages to a rather widespread phenomenon that is at the heart of theoretical
linguistics. The theoretical interest of the wider linguistic community in the phenomenon was given a
major impetus by the publication of John Haiman & Pam Munro’s seminal book Switch reference and
universal grammar in 1983, the second volume in the Typological Studies in Language in which this
book is also published. Their often cited definition of SR can serve as the starting point for this
volume:
Canonical SR is an inflectional category of the verb, which indicates whether or
not its subject is identical with the subject of some other verb (Haiman & Munro
1983b: ix).
For example, the Papuan language Usan uses a set of suffixes on the medial verb su ‘cut’ to indicate
that the subject of ‘cut’ is the same -ab ‘SS’ (1) or different -ine ‘DS’ (2) from the subject of the final
verb ‘go down’.2
(1) Usan [TRANS-NEW GUINEA], Reesink 1983: 217-8
a. ye nam su-ab isomei
1SG tree cut-IDENT I.went.down
‘I cut the tree and went down.’
b. ye nam su-ine isorei
1SG tree cut-NONID it.went.down
I cut the tree down.
However, almost all key aspects of Haiman & Munro’s definition have been challenged in one way or
another.3 For instance, there are languages whose SR system is not a verbal inflectional category, but
rather marked by opposed (sets of) free conjunction markers, as in the Brazilian language Maxakalí,
exemplified in (2).
(2) Maxakalí [MACRO-JÊ], Popovich 1986: 355, cited in Rodrigues 1999: 197-8
a. ʔĩ-mõŋ tɨ ʔ-nĩn
3-go and.IDENT 3-come
‘He went and returned.’
b. ʔĩ-mõŋ ha ʔ-nĩn
1 Jacobsen (1967) was based on a presentation given in 1961, so we can say that in fact the term switch reference
is over half a century old. 2 For ease of comparison, we gloss opposed SR values as IDENT (identity) versus NONID (nonidentity) in all
examples of this chapter. 3 Haiman and Munro were certainly aware of most of these challenges, which is why they include the word
‘canonical’ in their definition.
2
3-go and.NONID 3-come
‘Hei went and hej returned.’
Perhaps most famously, the subject requirement has been challenged, for instance on the basis of
‘mismatches’ such as in (3), from the Californian language Eastern Pomo (McLendon 1978, discussed
in Foley & Van Valin 1984 and Matić et al. 2014).
(3) Eastern Pomo [POMOAN], McLendon 1978, cited in Foley & Van Valin 1984: 119-120
a. há: káluhu-y si:má:mérqaki:hi
1SG.A go.home-IDENT went.to.bed
‘I went home and then went to bed.’
b. há: káluhu-qan mí:p si:má:mérqaki:hi
1SG.A go.home-NONID 3SG.A went.to.bed
‘I went home and he went to bed.’
c. há: xá: qákki-qan wi qa:lál tá:la
1SG.A water bathe-NONID 1SG.U sick become
‘I took a bath and got sick.’
Judging from (3c), semantic role rather than subjecthood ultimately determined the pivot of the SR
system in Eastern Pomo. In fact, the SR systems of some languages suggest that SR marking goes well
beyond referential identity, or perhaps rather may be overruled by non-referential factors to do more
generally with discourse cohesion. This is shown in the following example from the Ecuadorian
language Tsafiki.
(4) Tsafiki [BARBACOAN], Dickinson 2002: 137
junni man=ja-na-sa wata=te aman chide la-ri-bi
then again=come-PROG-NONID year=LOC now bone come.out-CAUS-PURP
man=ji-man-ti-e
again=go-SIT-REP-DECL
‘They say then, coming back, after one year he went to take out the bones.’
All other things being equal, the same referent marker -to appears in situations of referential identity
between the subject participants of two clauses, and the different referent marker -sa when there is no
such referential identity. However, other factors, to do with discourse cohesion may overrule this basic
set-up. In (4) for instance, the fact that the events are not temporally adjacent causes the different
referent marker to appear, in spite of the referential identity between the subject participants.
Finally, even within systems where referential identity is at stake, the interpretation of “identity” is
subject to cross-linguistic variation. For instance Austin (1980) reports for the Australian language
Diyari, that coreferentiality mismatches may occur if (i) the number of the subordinate clause subject
exceeds the number of main clause subject, and (ii) the subordinate subject referent set includes the
main clause subject, as in (5).
(5) Diyari [PAMA-NYUNGAN], Austin 1980: 16.
ngathu nganyja-yi ngalda diyari yawada yathayatha-lha
I.ERG want-PRES we.DL.INCL.NOM Diyari language.ABS speak-IMPL.IDENT
‘I want us to talk Diyari.’
Examples such as the ones in (2)-(5) raise a number of questions about the unity and demarcation of
SR as a phenomenon. Nevertheless, a few points seem to be rather basic for SR systems.
I. SR systems encode mutually exclusive values whose semantics include referential (non-)identity
Many languages have some complex sentence construction that only allows for identical participants
of the clauses in the construction. What is special about SR systems is that, for the functional
equivalent of the identity construction, there is an equally constrained non-identity construction.
Although it has been made clear in by several authors (e.g. Reesink 1983, Roberts 1988, Stirling
3
1993), that SR systems may be sensitive to a number of other parameters, referential (non-identity) is
always an important part of what determines which construction must be employed.
II. SR systems track a particular, generalized role
If we focus on referential (non-)identity, the work of Foley & Van Valin (1984) and Van Valin &
LaPolla (1997) made clear that one of the defining characteristics of SR systems is that they do not
track a particular participant, but rather a role. This role may be a syntactic (typically subject),
semantic (typically agent) or pragmatic (typically topic) one, but it is defined in a general way,
applicable across many contexts.
III. The domain of application of SR is one level above the clause, below the entire discourse
A third characteristic of SR systems is their domain of application. They are narrow-domain
operations (Kibrik 2011), meaning that they operate on stretches smaller than the entire discourse.
Unlike reflexives, SR systems do not operate within clauses, but rather between clauses.4
IV. SR systems are morphologically marked
A final characteristic of SR systems is that they are associated by some overt morphological marking
for at least all but one of the opposed values. The type of morpheme that is used (free form, affix,
clitic, non-concatenative operation) is in principle orthogonal. This requirement excludes systems that
have obligatory gapping for identity relations and obligatory expression of arguments in the case of
non-identity relations.
Two additional comments are in order to place the above in an appropriate perspective. First, we
emphatically do not claim here that the four characteristics above should be taken to mean that switch
reference is (or indeed should be) clearly distinguished from other linguistic phenomena that deal with
similar issues to do with interclausal referential coherence. Switch reference systems may develop out
of or into other, similar subsystems. Studying systems that do not quite fit all of the above
characteristics is therefore instructive and highly recommended.
Second, SR is not just limited to isolated sentences. In many languages where it is found, it is
integral to cohesion and text interpretation. It can be used to form clause chains of high complexity or
length that are “pervasive” in day-to-day language use (Crowley 1998: 247).Therefore, any description
of such languages relies heavily on a clear understanding of the grammar, function and historical
development of SR systems in general, and consequently, for a full understanding of the phenomenon
of SR, we need to refer to all of these dimensions.
The present volume is intended as an exploration into some of the questions relating to SR and their
possible answers. In particular, the following questions are addressed.
1. What is SR (and what is it not)?
Given the many exceptions and/or expansions to Haiman & Munro’s definition given above, the
question becomes relevant whether SR can be defined as a unified phenomenon. Moreover, the
opposite question (What is not SR?) is equally problematic, given the presence of similar linguistic
phenomena like logophoricity, gapping, long-distance reflexivity, converbs, obviation etc. This issue
is explored in more detail in Section 3.1.
2. What is the cross-linguistic variation space of SR systems
4 Although we acknowledge that ‘clause’ and ‘sentence’ are difficult concepts to define in a cross-linguistically
salient and consistent way, it is usually possible to do so within individual languages. Nevertheless, there is still
a wide range of constructions that fall under this definition, from (almost) monoclausal converbal constructions
to paragraph-length clause chains.
4
Section 3.2 will address this related question. What are the relevant variables of cross-linguistic
diversity within SR systems, and what are their values?
3. What is the function of SR?
Authors working from a variety of theoretical perspectives have proposed different functions for SR
and consequently disagreement has arisen about the proper theoretical embedding of SR, ranging from
syntactic binding to discourse-cohesion. These different perspectives are discussed in Section 4.
4. How does SR evolve?
Given the often contiguous geographical areas in which SR occurs, it is tempting to assume that SR
spreads through contact. However, as mentioned above, the phenomenon is rather widespread, and
occurs in many unrelated parts of the world. Moreover, even in contiguous geographical areas the SR
systems in different languages or families may be rather different from each other. This raises
questions about the diachronic stability and propensity to diffuse through contact of SR. Moreover the
geographical spread of the phenomenon may warrant alternative explanations, relating to e.g. human
cognition and communicative functionality. Issues relating to the diachronic development of SR
systems are discussed in Section 5.
Before moving on to the issues raised by questions 1-4, we start out with a brief historical overview of
the study of SR between 1967 and the present.
2. A brief history of switch reference
In a paper published in 1967, William H. Jacobsen introduced switch reference a new term to
linguistics, defining SR as a system where “a switch in subject or agent (…) is obligatorily indicated in
certain situations by a morpheme, usually suffixed, which may or may not carry other meanings in
addition” (Jacobsen 1967: 240). He discusses data of three North American languages, Tonkawa,
Kashaya, and Washo, of which some authors had proposed a distant genetic connection to the Hokan
family.5 One of the points addressed by Jacobsen in this paper is whether or not the presence of a
switch-reference system can be an argument in discussions on distant genetic relationships, a question
which he answers negatively. He briefly contemplates areal diffusion in southwestern USA as an
explanation for the patterns found, but finds the evidence for that equally indecisive.
The main point of Jacobsen’s paper, however, is to describe certain important variables within
switch-reference systems in order to give an idea of the typological variation that may be encountered.
He discusses the position of the SR marker, markedness of the opposed values, the syntactic/semantic
contexts in which SR marking occurs, the nature of identity (discussing person matches but number
mismatches), the relative order of the marked and controller clauses, additional meanings carried by
SR markers, nominalizing effects of some SR markers. He moreover compares SR to similar devices
like reflexivity, obviation, and fourth person systems. In short: many relevant typological parameters
of SR systems (see Section 3.2) were already included in this remarkable paper.
Publications subsequent to Jacobsen’s paper in the 1970s made it clear that the geographical extent
of SR was considerable. Much more information on SR systems in North America came to light (e.g.
Munro 1976, Oswalt 1976, Watkins 1976, Winter 1976, Langdon & Munro 1979) but also in other
areas, in particular Papua New Guinea (Longacre 1972, Scott 1973, Huisman 1973, Olsen 1978,
Haiman 1979).
This increase in information gave rise to two edited volumes with SR as their topics in the
beginning of the 1980s: Munro (1980a) and Haiman & Munro (1983a). Munro (1980a) included an
overview study of SR in Australian languages (Austin 1980), as well as contributions on South
American languages (Weber 1980, Cole 1983 on Quechuan languages, Longacre 1983 on a
Tucoanoan language in comparison with a Papuan language), non-canonical systems from the
Caucasian area (Nichols 1983) and from Yup’ik Eskimo (T. Payne 1980, Woodbury 1983).
5 The Hokan hypothesis has waned over the years in the absence of convincing evidence. Today, Tonkawa and
Washo are considered to be language isolates, and Pomoan languages (Kashaya is Pomoan) are presently not
considered to be part of the Hokan family (see Campbell 1997: 290-305 for a historical overview and
assessment).
5
Apart from increasing our knowledge of types of SR systems and their geographical distribution,
these two collections can be thought of as pivotal for the study of switch reference in the sense that
they introduced the phenomenon to a much broader audience of linguists and - perhaps even more
importantly - because they have outlined the issues that are relevant to the broader study of linguistics.
The issues raised in these two volumes are still of importance in the study of switch reference today
(although they have not received equal attention) and they will be of relevance to the present volume
as well. In fact, the structure of this paper is based on what we think are the most important themes of
Munro (1980a) and Haiman & Munro (1983a):
(i) The typology of SR
This includes the delimitation of the phenomenon vis-à-vis other, related phenomena (e.g.
Munro 1980b, Comrie 1983, Givón 1983, Haiman 1983, Heath 1983, see also Nichols 1983) as
well as the variables that describe cross-linguistic variation between SR systems in different
languages (D. Payne 1980, Munro 1980c, 1983, Cole 1983). This will be the topic of Section 3.
(ii) The function of SR
Haiman & Munro (1983b) famously declared that reference tracking was the main functional
motivation for languages to develop a SR system. This claim has been challenged in later
publications, to which we come back in Section 4.
(iii) The diachronic development of SR
Two major lines of inquiry can be distinguished with respect to this topic. The first is based on
the language-internal development of SR systems, in particular the diachronic sources of SR
markers or constructions (e.g. Givón 1983, Haiman 1983, Jacobsen 1983) and a second that
focuses on the contact-induced diffusability of SR systems, mainly instigated by the often
contiguous areas where SR systems occur (e.g. Austin 1980, Jacobsen 1983, Longacre 1983) .
Section 5 deals with this third theme.
The theoretical aspects of Munro (1980a) and Haiman & Munro (1983a) opened up a different
debate, placing SR rather in the center of attention of linguistic theories, such as Government &
Binding (Finer 1984, 1985), the Minimalist Program (Camacho 2010, McKenzie 2012, Keine 2013),
Role & Reference Grammar (Foley & Van Valin 1984, Van Valin & LaPolla 1997), as well as more
discourse-semantic oriented approaches such as Discourse Representation Theory (Stirling 1993),
Neo-Gricean pragmatic theory (Huang 2000), and cognitive-discourse analytical approaches (Kibrik
2011). These approaches to SR will be discussed in more detail in Section 4.
The publication of Munro (1980a) and Haiman & Munro (1983a) also changed the nature of the
descriptive studies that appeared at later occasions, as they formed points of reference. Later
descriptive studies, therefore, were much more theoretically informed, shifting the emphasis to
theoretical challenges posed by individual languages (see e.g. Reesink 1983, Roberts 1988, Steward
1988, Mithun 1993, O’Connor 1993, Watkins 1993 to mention a few, as well as most of the
descriptive studies in Munro 1980a and Haiman & Munro 1983a themselves).
3. The dimensions of SR
As mentioned above, SR is not a highly unified phenomenon. This has two consequences relevant for
the study of SR. First, it is not always easy to draw the line between SR and other linguistic
phenomena. Second, it makes the study of typological variation between SR systems necessary. In this
section we address these two questions, starting with the external boundaries of the domain of SR.
3.1. The outer dimensions: SR vis-à-vis other phenomena
Switch-reference is not a phenomenon that exists in a vacuum. It may develop out of or into
constructions or grammatical systems that are functionally related to it. In the first place, most SR
systems are analyzed at least partially as reference-tracking systems. This makes switch reference part
of a family of constructions that deal with signaling referential (dis)continuity, in particular those
constructions that deal with referential (non-)identity between two (or more) clauses. In this section
we discuss a number of these related construction types and indicate where they prototypically differ
from or pattern with (canonical) SR. As will become clear, it is impossible to strictly separate SR from
6
other constructions because of the existence of non-canonical instances of construction types that blur
the boundaries. In fact, it is undesirable to draw a hard line between the different construction types,
because their comparison may shed light on diachronic pathways that may give rise to SR systems, or
that may describe the transition of an SR system into another. We will come back to the diachronic
development of SR systems in section 5.
Gapping of arguments that are coreferential with some argument in another clause are examples of
reduction strategies. Like SR, the notion of pivot is often important for reduction strategies, in
particular for gapping. Whereas the pivot for gapping in English coordination is the subject, in Dyirbal
it is the absolutive argument.6
(6) Dyirbal [PAMA-NYUNGAN], Dixon 1979: 62
ŋuma yabu-ŋgu buṛa-n banaga-nyu
father mother-ERG see-NONFUT return-NONFUT
‘Motheri saw fatherj and Øj returned.’
Especially systems where gapping is obligatory for coreferential and prohibited for non-coreferential
arguments between two clauses come very close to SR systems, for example the complements of
want-constructions in English.
(7) English [GERMANIC]
a. I want Ø/*me to help you
b. I want *Ø/John to help me
In fact, the main difference between these grammaticalized gapping rules and SR is the fact that the
former requires some additional marking (see Matić et al. 2014).7 Nevertheless, as Jendraschek (this
volume) argues, the system of Iatmul can be analyzed as a SR system where the difference between
same-subject and different-subject verbs is the fact that the former does not have any person marking
and the latter does.8
The situation is slightly more complex for the comparison with converbs. A converb is defined as
“a nonfinite verb form whose main function is to mark adverbial subordination” (Haspelmath 1995:
3). Converbs are often (but not necessarily) restricted in their usage by referential identity. Haspelmath
sets up the following typology based on the behavior with respect to subject (co-)reference (Ibid. 10,
italics in original):
Table 1. Subject reference in converbs (Haspelmath 1995:10)
same subject different subject varying subject
implicit-subject converb typical unusual unusual
explicit-subject converb unusual typical unusual
free-subject converb unusual unusual typical
In Table 1 a distinction is made between those converbs that allow (or require) an explicit subject and
those that do not. On the basis of those behavioral patterns, the converbs that do not take an explicit
subject are associated with same-subject environments, while those that have an explicit subject are
typically associated with different-subject contexts.
6 Nevertheless, as Dixon (1979: 63) mentions, many languages that are morphologically ergative still have
accusative syntactic rules (like for gapping). This is another parallel with SR, where many systems seem to have
a subject (S/A) pivot, regardless of their morphological alignment patterns. 7 Another, more gradual difference is that, in gapping, it tend to be the grammatically more tight constructions
(like complementation or relativization) that have the most grammaticalized and least flexible gapping rules,
whereas SR tends to occur more often in looser grammatical structures such as adverbial clauses than in
complementation or relativization. Another gradual difference is that gapping is often more flexible than SR in
the distribution of the referentially dependent element (see Matić et al. 2014: 26-27). 8 The same-subject verb forms are also marked for relative tense with markers that are not found in different-
subject environments. One could also argue that these in fact mark the SR system. In that case it is the coreferent
situation that is more marked than the non-coreferent one.
7
In this sense, it seems that it is gapping rather than the converb form that is responsible for the
reference tracking behavior in these constructions. However, the situation is not that simple. As
Nedjalkov (1995) shows, same-subject converbs and different-subject converbs may differ from each
other morphologically. For instance, one of the two may carry an additional morphological marker, as
is the case for the language isolate Nivkh, soken in eastern Russia, where different subject converbs
include an additional suffix -gu (Ibid. 112). Alternatively, the same-subject and different-subject
converbs may both be morphologically marked by unrelated inflectional morphemes, like in Hopi.
(8) Hopi [UTO-AZTECAN], Kalectaca 1978: 149-150, cited in Nedjalkov 1995: 114
a. nu’ pakí-t pu’ qatuvtu
I come.IDENT.CONV then sit.down
‘I came and sat down.’
b. nu’ pakí-q pu’ pam qatuvtu
I come.NONID.CONV then she sit.down
‘I came and sat down.’
Because finiteness is orthogonal to SR (see Section 3.2) the converb construction type and SR in cases
like Nivkh and Hopi overlap. Similar overlaps may exist between SR and constructions that are akin to
converbs, like participial constructions and medial verbs: if they morphologically mark am opposition
between referential identity and non-identity, they are to be analyzed as instances of SR.
One type of system usually discussed under the heading of switch-reference involving medial verbs
does not seem to involve any additional marking at all: the so-called implicit chain in the Tucanoan
language Kotiria, also called Guanano (Waltz 1976 Longacre 1983) and in the Chibchan language
Tunebo (Headland & Levinsohn 1977). For these languages it is claimed that a combination of same
subject and temporal succession as well as of different subject and simultaneity are unmarked in the
language.
(9) Kotiria (Guanano) [TUCANOAN], Waltz 1976: 25 and 38, respectively
a. waha süre
go.MED arrive.PST
‘Having gone, he arrived there.’
b. tina chü tiro cjuri ñaca taha jüna
they eat.MED he turtle character come.PST finally
‘While they were eating, that turtle character finally came.’
Stenzel (this volume) casts doubt on whether this is in fact a grammaticalized pattern in Guanano.
However, it is a conceivable pattern that raises the question whether such a system should be classified
as an SR system or rather as an instance of a reduction strategy.
There is a group of strategies that exhibit explicit morphological marking, which also come close to
SR, and in fact sometimes considered to be a (non-canonical) subtype of SR. We discuss them in turn,
spending a little more time on logophoricity on the basis of discussions by Comrie (1983) and Stirling
(1993) because similar issues recur for the other construction types. Logophoricity shares a number of
features with switch reference. In Igbo, the subjects of complements of verbs of communication are
marked for coreferentiality with the source of the information.
(10) Igbo [NIGER-CONGO], Comrie 1983: 21
a. ó sìrì nà ó byàrà
he said that he came
‘Hei said that hej came.’
b. ó sìrì nà yá byàrà
he said that LOG came
‘Hei said that hei came.’
A special pronominal form is used when the subject of the complement clause is coreferential with the
agent of the communication act. This is a grammaticalized system in that the use of the wrong
pronominal form necessarily leads to a different interpretation.
8
There are certainly also differences between SR and logophoricity (see discussions by Comrie 1983
and Stirling 1993), but these differences may sometimes be blurred by the existence of languages with
characteristics of both types of systems. First, whereas SR is most prototypically marked on the verb,
or alternatively on conjunctions, logophoricity is prototypically marked on the (pro)noun. However,
this is not always true. Comrie (1983) reports on the Nigerian language Gokana, which displays a
system with many logophoric characteristics, where coreferentiality is marked on the verb.
(11) Gokana [OGONI], Comrie 1983: 21
a. aè kɔ aè dɔ
he said he fell
‘Hei said that hej fell.’
b. aè kɔ aè dɔ-ɛ
he said he fell-LOG
‘Hei said that hei fell.’
At the other end, there are SR systems where, although marked on the verb, the marking patterns are
merged with the bound pronominal markers. In particular for pro-drop languages, these bound
pronominal markers may be argued to constitute pronouns rather than inflection (see e.g. Siewierska
1999, Corbett 2006). An example of a language where the system of bound pronouns and SR marking
are merged is Kobon, whose SR paradigm is given in Table 2 (Comrie 1983:20).
Table 2: SR suffixes in Kobon [TRANS-NEW-GUINEA]
identity non-identity
1SG -em -nö
2SG -(m)ön -ö
3SG -öm -ö
1DL -ul -lo
2/3DL -mil -lö
1PL -un -no
2PL -mim -be/-pe
3PL -öm -lö
The situation in Kobon in terms of the marking strategy is similar to what we find in fourth-person
systems, where coreferentiality is fused with the pronominal system. The situation in echo-referent
systems is rather more like canonical SR in that the marking of whether a dependent clause subject is
coreferent or not with the subject of a marked clause is not fused with person/number coreference, it
rather replaces it.
Munro (1980b: 2), reporting on the results of a conference on switch reference notes that at least
two characteristics set SR apart from other types of interclausal reference marking strategies:
1) In a "real" switch-reference system the same/different subject distinction is pervasive, not
restricted to a few constructions.
2) Switch-reference continues to operate even when no one would question the difference of two
subjects (i.e. between a first and third person subject)
These two characteristics set canonical SR further apart from canonical logophoricity. First whereas
SR systems are especially common in different types of adverbial clauses and may have extensions to
coordinate, complement, and relative constructions, logophoric systems are usually found in a much
more restricted set of contexts, namely complements of communication verbs. With respect to
Munro’s second point, logophoric systems prototypically apply only to third persons, non-third person
pronouns have no logohoric counterpart.
9
Furthermore, SR differs from logophoricity in the way the controller is defined. In logophoric
systems the controller is semantically defined as the source of the utterance (or thought or emotion),
independently from the syntactic role it has.9
(12) Gokana [OGONI], Comrie 1983: 32
mm dã ɛ gã kɔ aè dɔ-ɛ
I heard him mouth that he fell-LOG
‘I heard from himi that hei fell.’
SR systems have been described with a semantic pivot, but these semantic systems are still more
general than the very specific semantic role for the controllers of logophoric pronouns (see e.g. Eastern
Pomo, discussed above).
Finally, there is a difference in tendency between SR ad logophoricity in that the former, if there is
a marked and unmarked feature value, it generally is the marker for non-identity that is marked. In
logophoric systems this is reversed. Coreferent pronouns are usually the marked ones and non-
coreferent third person pronouns are identical to third person pronouns that appear in independent
clauses.
Long-distance reflexivity also involves the explicit marking of interclausal (non-)coreference, by
the use of a reflexive form of one of the arguments in the subordinate clause. In (13), the reflexive
element ziji may refer back to all three possible antecedents.
(13) Mandarin [SINO-TIBETAN], Cole et al. 2001: xiv
Zhangsan renwei Lisi zhidao Wangwu xihuan ziji
Zhangsan think Lisi know Wangwu like self
‘Zhangsani thinks Lisij knows Wangwuk likes selfijk.’
Some of the criteria that set logophoricity apart from SR are also valid to argue for a distinction
between long-distance reflexivity and SR. Like logophoric constructions, long-distance reflexivity is
often marked in the (pro)nominal domain to the extent that the reflexive markers share with
(pro)nouns the ability to function as an argument. Moreover, in the cases where a morphological
difference is made,10 the coreferent rather than the non-coreferent is morphologically marked.
Furthermore, long-distance reflexives tend to occur in complement clauses (specifically infinitive or
subjunctive clauses) whereas SR seems to prefer ‘flatter’ syntactic structures. Another difference from
SR is that, whereas the controller of LDR is prototypically a subject (see Cole et al.) but the reflexive
pronoun itself often exhibits more freedom. It can be the object as in (13) above, but also a subject, as
in (14), a dative/benefactive, as in (15). or a possessor, as in (16).
(14) Korean [ISOLATE], Lee 2003: 436
Marikoi-ga zibuni-ga ichiban moteru-to shinjiteiru
Mariko-NOM LDR-NOM best be.popular-COMP believe
‘Marikoi believes that selfi is the most popular.’
9 In Gokana, the pivot is also rather free, compare (11b) above with a subject pivot above to these two sentences
with an object and possessor pivot, respectively.
a. aè kɔ oò div-èè e
he said you hit-LOG him
‘Hei said that you hit himi.’
b. aè kɔ oò ziv-èè a gĩã
he said you stole-LOG his yams
‘Hei said that you stole hisi yams’
Nevertheless, this is not a typical situation for logophoric systems. 10 For most cases it is questionable whether one can really speak of a difference in morphological markedness for
long distance reflexives, since they tend to be morphological roots, just like non-reflexive pronouns.
Nevertheless, since long-distance reflexives are often restricted to specific types of subordinate clauses,
compared to the main clause the reflexives are marked. Moreover, when there is morphological marking, like in
Malay (see example 15) and Turkish (Rudnev 2008), it appears on the long-distance reflexive.
10
(15) Malay [AUSTRONESIAN], Cole et al. 2001: xvii
Ahmadi tahu Salmahj akan membeli baju untuk diri-nyai/j/k
Ahmad know Salmah will buy clothes for LDR-3SG
'Ahmad knows Salmah will buy clothes for him/herself.'
(16) Mandarin [SINO-TIBETAN], Huang & Tang 1991: 265
Ziji de xiaohai mei de jiang de xiaoxi shi Lisi hen nanguo
LDR of child not get prize of news make Lisi very sad
‘The news that his own child did not get the prize made Lisi very sad.’
A final difference is that whereas LDR can be bound within its clausal domain, as shown by the three-
way ambiguity of (13) this is not true for SR. In fact, long-distance binding seems to be rather avoided
in some circumstances. In (17) the person features of the potential antecedents (the subjects) do not
match, and consequently the potential antecedents are reduced to the subject of the local domain. The
non-matching intermediate subject is not a potential antecedent because of the feature mismatch, but it
also blocks the highest subject as a possible antecedent.
(17) Mandarin [SINO-TIBETAN], Cole et al. 2001: xiv
Zhangsani renwei woj zhidao Wangwuk xihuan ziji*i/*j/k
Zhangsan thinks I know Wangwu likes self
' Zhangsan thinks I know Wangwu likes self.'
On the other hand, there are also examples of long-distance reflexives that require or prefer non-local
antecedents, as in Danish.
(18) Danish [INDO-EUROPEAN, GERMANIC], Testelets 2008, cited in Rudnev 2008
a. Peteri hørte Anne omtale sigi
Peter heard Anne mention self
‘Peter heard Anne mention him.’
b. *Peteri fortalte Michael om sigi
Peter told Michael about self
‘*Peter told Michael about himself.’
Fourth-person systems have been regarded as non-canonical instances of SR (see e.g. Munro
1980b, Haiman & Munro 1983b). The term refers to a cross-referencing paradigm that contains a form
which cross-references a third person participant which is coreferent with a participant that occurs
elsewhere in the construction. An example is given in (19).
(19) Central Yup’ik [ESKIMO-ALEUT], T. Payne 1980: 65.
a. Doris-aq quya-u-q Tom-aq cinga-llra-ø-ku
Doris-ABS happy-INTR-3SG Tom-ABS kiss-because-3SG:SUBJ-3SG:OBJ
‘Dorisi is happy because shej kissed Tom.’
b. Doris-aq quya-u-q Tom-aq cinga-llra-mi-ku
Doris-ABS happy-INTR-3SG Tom-ABS kiss-because-3SG.COREF:SUBJ-3SG:OBJ
‘Dorisi is happy because shei kissed Tom.’
The only difference between (19a) and (19b) is the presence of the suffix -mi in the latter, which forces
the coreferent interpretation, whereas its absence renders a coreference reading impossible. A few
differences with canonical switch reference are immediately clear: fourth persons are marked within
the (bound) pronominal paradigm (though this is not a decisive difference), the coreferent rather than
the non-coreferent form is marked, and the coreference marking is restricted to third person referents.
The marking of coreference for third persons in complex sentences is restricted to certain adverbial
subordinate (connective - T. Payne 1980: 68, oblique - Woodbury 1983) moods and the appositional
mood, though in the latter case there is no opposed non-coreferent construction because appositional
11
mood clauses always have a coreferent subject (Woodbury 1983).11 In its usage in the adverbial
clauses the system is functionally akin to canonical SR, though with the differences noted above.
However, coreference marking extends to possessors as well.
(20) Central Yup’ik [ESKIMO-ALEUT], T. Payne 1980: 80
Yero-q angya-mi-ni qava-llru-u-q.
Yero-ABS boat-3SG.COREF-LOC sleep-PST-INTR-3SG
'Yeroi fell asleep in hisi boat.’
This is unlike SR, and reminiscent of the long-distance reflexives discussed above. However, they are
different from each other in several respects. First, long-distance reflexives are often free,
monomorphemic forms acting as nominal and fourth person markers are part of an inflectional
paradigm of person markers. Second, whereas long-distance reflexives tend to occur in complement
clauses rather than adverbial clauses, fourth person markers display the opposite pattern. Finally, the
interpretation of the fourth person marker in adverbial clauses is that it is coreferent with the subject of
the directly superordinate clause, in long-distance-reflexive constructions there are usually more
alternative antecedents.
Although fourth person systems are generally associated with languages of the Eskimo and Yup’ik
branches of the Eskimo-Aleut family, some languages in the Tupian language family of South
America display a remarkably similar pattern. In the systems of both these families, there is a special
form for (at least) the coreferential third person, and this pattern extends to the marking of bound
possessor pronouns. A contrasting example of the interclausal type is given for the Tupian language
Karo.
(21) Karo, [TUPIAN, RAMARAMA], Gabas Jr. 1999: 200-201
a. péŋ yaʔwat-t [to=wé-a kanãp]
white.man leave-IND 3COREF=cry-GER when
‘The white mani left when hei cried.’
b. o=yaʔwat-t [aʔ=ket-a kanãp]
I=leave-IND 3SG=sleep-GER when
‘I left when he slept.’
The state of the coreferentiality markers in the different Tupian languages is quite diverse. Some
languages have lost the dimension altogether, others have maintained it for the third person only,
and/or mostly restricted to verbs, and yet other languages have a full set of coreferential prefixes
occurring in both verbal an nominal contexts, like Tocantins Asurini. Jensen (1997) argues that the
system for proto-Tupian was like the one found in Tocantins Asurini, and has decayed in other
languages. This makes the system found in Tupian languages slightly more like prototypical SR. The
reconstructed coreference markers for proto-Tupí-Guaraní (the largest subbranch of Tupian) are given
in Table 3 (Jensen 1997).
Table 3: reconstructed coreference forms of proto-Tupí-Guaraní
1SG *wi-
2SG *e-
3 *o-
1INCL *jere-
1EXCL *oro-
2PL *peje-
The fourth-person systems as discussed in the previous paragraphs are in a number of ways
reminiscent of systems of obviation. In fact, the same term ‘fourth person’ is sometimes used to refer
11 Verbs in the appositive mood only cross-reference the absolutive argument, while the coreference marking
functions to a large extent on a subject pivot. The upshot of this is that coreference in appositional mood clauses
is only marked for intransitive verbs (Woodbury 1983). Note that this analysis differs from the one offered in T.
Payne (1980) who argues for fourth person object markers.
12
to these systems as well. In obviative systems the values of the feature person can also be said to
include two types of third person: an obviative and a proximate type. In Ojibwa this is, among other
things, indicated by pronominal affixes on the verb.
(22) Ojibwa (central dialect) [ALGONQUIAN], Mithun 1999: 76
ogii-biindoomon-an iniw anishnaabe-n ahaw misaabe
3-carry.in.garment-3OBV those.OBV people-OBV that giant
‘Giant (proximate) carried people (obviative) in the fold of his garment.’
Although, like SR, obviation contributes to reference tracking, there are a number of clear differences
as well. First, obviation, in Algonquian languages at least, is marked on nouns and demonstratives as
well, whereas one of the special things about SR is its ‘misplaced’ marking on the verb rather than on
the referential elements themselves (see Haiman 1983). Obviation does not necessarily apply to
complex sentences. As can be seen in (24) obviation is marked for simple sentences as well.
Moreover, prototypically, proximate and obviate status of participant is a discourse-level matter
entirely, determined by the relative centrality of the participants in the discourse. Nevertheless, a
recurring critique on Haiman & Munro’s (1983b) syntactic definition of SR is that it downplays the
discourse dimensions of SR while in fact there is abundant evidence of the interaction between larger
discourse structure and SR. We come back to the discourse perspective on SR in Section 4.3.
Echo-referent or echo-subject systems, finally, seem to be confined to a group of Oceanic
languages in Southern Vanuatu. These systems have a single special form for coreferential arguments
that appears in clause chains, which replaces person cross-reference markers. The coreferentiality
marker in these systems simply indicates that the subject of the reference clause should be copied to
the marked clause. Non-coreferential clauses are encoded in the same way an independent clause
would in terms of person marking. This is shown by the following pair of examples from Lenakel.
(23) Lenakel [AUSTRONESIAN], Lynch 1983: 211
a. i-ɨm-vɨn (kani) r-ɨm-apul
1EXCL-PST-go (and) 3SG-PST-sleep
‘I went and he slept.’
b. i-ɨm-vɨn (kani) m-ɨm-apul
1EXCL-PST-go (and) ES-PST-sleep
‘I went and he slept.’
In structural terms, echo-referent systems differ from canonical SR in the sense that it is the coreferent
rather than the non-coreferent situation that is morphologically flagged. Another relatively uncommon
characteristic in comparison to most known cases of canonical SR is that the coreference marker is
prefixed to the verb stem, and the reference clause generally precedes the marked clause.12 Moreover,
number marking is separate from coreferentiality marking, but number is taken into account when
determining whether a participant is coreferent or not. Finally, although the pivot in the marked clause
is always the subject argument, some flexibility with respect to the antecedent is possible. When no
ambiguities arise the echo-referent marker may also have the object as its antecedent Lynch (1983:
215-6). This is, however, not true for the related language Whitesands (Hammond 2014: 276). Apart
from these differences, echo-referent systems also share a number of the hallmarks of SR, especially
the systems found in Papuan languages. Like in those systems, echo-referent marking is found in
syntactically ‘flat’ chaining constructions that may stretch over large chunks of narrative, ignoring
daughter-subordination constructions like relativization and complementation (Hammond 2014).
Even on the basis of this brief and non-exhaustive overview it is clear that it is rather difficult to
clearly distinguish SR from other types of linguistic phenomena that share certain functionalities.
Perhaps one can distinguish prototypical instances of these different phenomena, but the existence of
non-canonical, intermediate linguistic systems prevents a clear delimitation. It is also not our intention
to draw any sharp borders around SR. In fact, the fuzzy boundaries highlight the importance of
considering a wider range of constructions, because that will further our understanding of how SR
12 This fact was presented as correlating cross-linguistically with the prefixing of SR markers in coordinated
clauses in Haiman & Munro (1983b).
13
systems evolve, and what they can evolve into. This book, therefore, contains a number of descriptions
of systems that some people would hesitate to call SR or at most they would be classified as non-
canonical instances of SR. We come back to the diachronic aspect of SR systems in Section 5, and we
discuss the variation between different SR systems in the next section.
3.2. The internal dimensions of SR: typological variables and their values
In the opening Section, we indicated four basic contour principles of (canonical) SR. They are
repeated for convenience:
1. SR systems encode mutually exclusive (paradigmatically opposed) values whose semantics include
referential identity versus non-identity
2. SR systems track a particular, generalized role
3. The domain of application of SR is one level above the clause, below the entire discourse
4. SR systems are morphologically marked
It has often been observed that SR is a multivariate phenomenon (see e.g. Haiman & Munro 1983b,
Comrie 1983, Stirling 1993, Bickel 2010, see also e.g. the contributions to this volume by van Gijn
and Hill. In this subsection we try to give an overview of the most important variables and their
possible values, thus creating infrastructure for a refined comparison between SR systems. We start by
giving the basic ingredients of a SR construction on the basis of the four principles given above in
Figure 1.
Figure 1: The sub-parts of an SR construction
Before going into the sub-parts and the variables they introduce, we dwell briefly on terminology.
Following Munro (1980b) and Haiman & Munro (1983b) we distinguish between marked clauses and
reference clauses. Marked clauses are associated with an SR marking pattern, indicating
interpretational dependence on some other clause, which we term the reference clause. However, a
reference clause may in turn also be a marked clause, for instance in a chain of clauses. Therefore we
think it is useful to additionally distinguish the anchor clause, which is a reference clause but not a
marked clause by definition, as it is the finite clause of the construction where all dependencies find
their resolution. The “marker” in Figure 1 is meant in a broad sense: it can be an affix, a free form, a
non-linear operation, and also a zero realization. We broadly distinguish identity markers and non-
identity markers, which may have quite an extended functionality, but among their functions must be
that they mark referential (dis)continuity between the relative participant and the controller participant.
For the same reason as we distinguish reference clauses and anchor clauses, we distinguish between
controller participants and anchor participants. Finally, of importance are the event parameters, which
include aspects like place, time, reality status etc. of the events. It has been shown that these
parameters are of importance for distinguishing between identity and non-identity situations in at least
domain
Marked clause Reference clause
Marker
relative participant controller participant
context
relation
relation
Event parameters Event parameters relation
14
some SR systems around the world. We come back to these parameters in section 4.3 below and will
not discuss them in this section.
The context
On the widest level, the variable context refers to the situations in which SR is used. Although to our
knowledge, no systematic research has been carried out on this topic, it seems that SR is more
prototypically found in narratives rather than in conversations. That is, the expectation would be that,
if a language uses SR in conversations, it will probably also use it in narratives, though not necessarily
the other way around. The context in which SR occurs is also important to explain the intricate ways
in which SR interacts with the wider discourse structure, for instance when issues of referential
identity are overruled by other factors of discourse (see e.g. Stirling 1993), the interaction with e.g.
referential activation status and resolution of potential referential conflicts (see e.g. Kibrik 2011) and
the role of tail-head linkage (see e.g. De Vries 2005). We come back to some of these discourse factors
in more detail in the Section 4.3, and restrict ourselves here to the variables at the level of the domain
and below.
The domain
The domain of SR refers to the stretch of text required for the resolution of the dependent participant
reference. A SR construction prototypically contains at most one finite anchor clause, and at least one
marked clause. The domain for SR is the complex sentence, but the term ‘sentence’ masks the fact that
it may refer to a rather diverse set of constructions. One dimension that seems to be rather important
for SR is the opposition of clause chains versus (predominantly) biclausal structures. In particular
Papuan languages have a preference for long, paragraph-level chains that contain a number of marked
clauses before the sequence is closed off with an anchor clause (the chain may also start with an
anchor clause followed by several marked clauses). In North America, on the other hand, as Jacobsen
(1983) reports, the structures seem to be predominantly biclausal, and in South America, both types
occur. This variable is rather hard to capture formally, because in chaining languages, shorter chains
may occur, and in the more biclausally inclined languages, longer clause combinations may occur.
There is no natural cut-off point beyond which one would speak of a chain. Nevertheless, chaining
constructions raise a number of issues that do not arise (or not as prominently) in biclausal structures,
in particular the fact that distinctions between the levels of the sentence, paragraph, and in some cases
even the entire discourse are blurred or even irrelevant in some chaining languages (Longacre 2007).13
Another issue, to which we come back shortly, is that in long chains necessarily consist of relatively
flat (non-hierarchical) structures, because it would reasonably asking too much of the human mind to
process multiple embeddings the length of those chains. This is in principle not a problem in biclausal
structures. Perhaps as a consequence, we see SR popping up in different types of environments in
North America (like e.g. in logical adverbial, complementation and relative clauses - see e.g. Jacobsen
1983).
There is a further issue that relates to clause chains. In longer chains, marked clauses may either
refer directly to the finite reference clause, whether or not adjacent to it, or it may refer to the adjacent
clause, whether or not it is the finite verb. This variable may be termed local versus long-distance (or
focal) controller scope. Some languages display both types of scope relations. Zariquiey (2011, see
also this volume) reports two types of SR constructions in the Peruvian Panoan language Kashibo-
Kakataibo, which differ from each other in this respect. Likewise, Weber (1980) proposes two
possible scope relations for SR controllers in Quechuan languages, either to the directly superordinate
clause (whether adjacent or not) or - in case of non-subordination - to the adjacent clause (though see
Steward 1988 for a critique). Another pattern that may occur in chaining languages is clause skipping.
In clause skipping, one of the subclauses is ignored by the SR system, illustrated in (24).
13 Longacre (2007) regards SR to be a defining criterion of a clause chain. However, chains do not have to be
marked for SR. In fact, it is possible to create a chain in English with participial forms. What makes the
combination clause chains with SR special, however, is that it allows for subject switches in chains, whereas
chains in English would have to be same-subject chains.
15
(24) Usan [NUCLEAR TRANS-NEW-GUINEA], Reesink 2014: 242
a. igam-a
stay-2/3SG.NONID
b. munon eng is ibi eng g-ab
man that descend feces that see-IDENT
c. to-at qiter asi
follow-IDENT lift.up.head.IDENT look.IDENT
d. [mani eng erer y-ab bug-ab igam-a]
snake that on.top curled.up-IDENT sit-IDENT stay-2/3SG.NONID
e. g-arei
see-3SG.REM.PST
‘It was (= the situation was such)a (and) the man went down, saw the fecesb (and) following
(the track) he looked upc (and) sawe the snake sitting curled up there (in the tree)d.’
The events in (24c) which have their own internal SR logic, are ignored (skipped) by the identity chain
in the entire structure in (24) which is resolved in the anchor clause in (24e).
Relations between marked and reference clause
Between the marked clause and the reference clause various different types of relations may exist. One
way to classify the relations between the clauses is in terms of nexus. There has been a lot of debate
about the distinction between coordination and subordination, and it is unclear whether we can easily
separate the two in all cases, although they do seem to form natural clusters (see Bickel 2010). Foley
& Van Valin (1984) argue for a third type of nexus that has properties of both coordination and
subordination called cosubordination. There is an overall preference for SR to appear in relatively flat
and loosely organized nexus types, in particular ad-subordinate and co-subordinate constructions.
There is also an explicit (and controversial) claim that SR does not appear in fully coordinate clauses
(see Finer 1984, 1985, see also Weisser 2012, this volume, see also Section 4.1 of this chapter). It is
not immediately clear why SR, even in non-chaining areas, is less common in daughter-subordination.
A further subclassification of the type of relationship between the marked clause and reference
clause is a semantic one. Different semantic characterizations of complex sentences are given in a
number of the chapters in Shopen (2007), such as different semantic types of coordination
(Haspelmath 2007), complementation (Noonan 2007), and adverbial relations (Thompson et al. 2007).
For relativization (Andrews 2007) we can distinguish between restrictive and non-restrictive clauses,
as well as between the different positions that are relativized. The major subtypes for coordination,
complementation (in terms of predicate types), adverbialization and relativization (in terms of
relativized arguments)14 given in the aforementioned sources.
Table 4: Semantic sub-classifications of major clause combination types
Coordination Complementation Adverbialization Relativization
Conjunctive Utterance Time Subject
Disjunctive Propositional attitude Location Direct object
Adversative Pretence Manner Indirect object
Commentative (factive) Purpose Oblique
Knowledge Reason Genitive
Fear Circumstantial Object of comparison
Desire Simultaneous
Manipulation Conditional
Modality Concessive
Achievement Substitutive
Phasality Additive
Perception Absolutive
Negation
Conjunction
14 This is, strictly speaking, not a semantic subclassification, but it is nevertheless taken up here because it does
represent different ways in which the relative clause relates to a main clause. The classification follows the
accessibility hierarchy (Keenan & Comrie 1977).
16
As mentioned above, SR is probably most common in adverbial clauses and within that group
arguably for temporal clauses, although no worldwide survey has ever been carried out to confirm this.
A final characterization of the relation between marked clause and reference clause is one of
linearity. Haiman & Munro (1983b) note that there seems to be a correlation between the position of
the SR marker (as a prefix or a suffix) and the order of clauses for coordinating relations. There seems
to be a general preference for iconic ordering especially in terms of time sequence, leading to marked
clause-reference clause order for temporal clauses, and the opposite pattern for purpose clauses. It also
seems to be true for most languages that they allow for some flexibility in linear patterns: although
there might be a dominant order, many languages allow for minority patterns of alternative orderings.
One of those alternative orderings is center embedding, a pattern where the marked clause is
positioned inside the reference clause
(25) Maricopa [YUMAN], Gordon 1983: 88
hat [‘ii-sh anoq-m] nym-aaham-m
dog [wood-SUBJ small-NONID] DEM+ASC-hit-ASP
‘He hit the dog with a small stick.’
The marked clause and the reference clause
Since marked clauses are typically dependent clauses, one set of variables relates to what form the
verb takes, in particular with regard to inflection. Can the verb of the marked clause take any inflection
that an independent verb can take, or is there a reduction in the number of categories that can be
marked? Some SR markers, moreover, seem to have a nominalizing effect (e.g. in Quechuan
languages - see Van Gijn, this volume, Floyd & Norcliffe this volume and Jivaroan languages - see
Gnerre, this volume, and Overall, this volume), so dependent verbs may also acquire nominal
characteristics in their inflection. Inflectional potential is a variable in general for dependent or
subordinate verbs (see e.g. Lehmann 1988, Cristofaro 2003, Malchukov 2006, Van Gijn &
Hammarström, forthc.) but of special interest from the perspective of SR is obviously whether and
how agreement with the relative participant is expressed. There are five major possibilities
Expression of relative participant agreement
i. With verbal inflection as in independent clause
ii. With nominal (possessor) inflection
iii. With a different paradigm than the one in the independent clause
iv. Not expressed, unlike independent clause
v. Not expressed, as in independent clause
However, since SR constructions may also be intertwined with other inflectional categories, e.g. tense,
modality, aspect, it may reduce or alter the inflectional potential in other areas than agreement as well.
Beyond inflection, the main other set of variables relates to the internal syntax of the marked clause
and the reference clause: in particular what the rules are for the expression of the NP argument whose
reference is being tracked. So given a minimal SR construction consisting of a marked clause and a
reference clause, we may postulate three values for the expression of the tracked argument in either
clause: obligatorily expressed, optionally expressed, or obligatorily gapped. In case of optional
expression, the conditions under which the argument may or may not be expressed should be made
explicit (e.g. not expressed under coreferentiality, not expressed when topical, not expressed when the
clause is final, etc.).
The relative and controller categories
Much has been said about the factors that trigger identity or non-identity markers. In Haiman &
Munro’s (1983b) assessment, the all-important category is the subject, but as mentioned above, this
claim has met with a lot of criticism. Givón (1983) argued that systems that track subjects started out
as systems that track topics. In a broader survey, Foley & Van Valin (1984) argued for the fact that
there are languages with semantic pivots, where agents rather than subjects are tracked, and for
17
languages that have a pragmatically-influenced syntactic pivotal category.15 An example of a semantic
relative-controller relation, sensitive to agentivity rather than subjecthood was given above in (3) for
Eastern Pomo. In (26), we give an example of a pragmatically influenced SR system from Barai
(Olson 1978, 1981), cited in Foley & Van Valin (1984).
(26) Barai [KOIARIAN], Foley & Van Valin 1984: 350
a. fu miane sak-i-na barone
3SG firestick bite-3SG-IDENT die
‘A firestick bit him and he died.’
b. miane ije fu sak-i-mo fu barone
firestick DEF 3SG bite-3SG-NONID 3SG die
‘The firestick bit him and he died.’
In the example from Barai the determining factor for either the identity or the non-identity marker to
appear is the referential status of miane ‘firestick’. In (26a) firestick is not topical and so the
coreference that is marked is between the topical object and the subject of the final clause. In (26b) on
the other hand, the firestick is topical and therefore it takes preference over the object in determining
SR marking, resulting in non-identity marking. A related but slightly different type of pragmatic
influence on the relation between the controller and relative participant is discussed by Cole (1983) for
Imbabura Quechua subjunctive clauses, where the marked clause carries an identity marker if the
subject of reference clause is non-referential and the subject of the marked clause a speech act
participant (a similar pattern is reported for Amele in Roberts 1988).
(27) Imbabura Quechua [QUECHUAN], Cole 1983: 6-7
ali-mi Ø/ñuka/kan/*pay Juzi-wan parla-ngapaj
good-EVID one/I/you/*he José-COM speak-IDENT
‘It is good that one/I/you/*he speaks with José.’
As was mentioned for Tsafiki in example (4) above and as will be argued for other languages in
Section 4.3 a number of SR systems are also sensitive to other event-related parameters such as unity
of place and/or time or thematic coherence between events, represented by the relation between event
parameters in Figure 1. It is probably fair to say that we are only scratching the surface of the ways in
which pragmatic or discursive elements may or may not influence the appearance of (non-)identity
markers. One of the goals of this volume is to increase our knowledge on this point.
Another important point to make in reference to the relative and controller categories is that we
should distinguish between the categorical restrictions for the anchor participant and those for the
relative participant, as they might differ from each other. This becomes especially clear from reports
on Panoan languages. Consider for instance the system of Matís (Ferreira 2005, identity markers only)
in Table 5.
Table 5: identity markers of Matis [PANOAN]
Marker Relative time relative argument reference argument
ash sequential S/A S
shun sequential S/A A
ak sequential P S/A
ek simultaneous S/A S
kin simultaneous S/A A
15 Stirling (1993) takes it one step further, claiming that SR is about event continuity, of which referential
continuity is a part, but it also appears from here comparative work that, whereas systems may differ from each
other in terms of the factors that influence the appearance of identity or non-identity markers, referential
(dis)continuity is present as a factor in all of the systems surveyed by her. Therefore we treated issues relating to
event continuity separately in relation to context, see above.
18
sho simultaneous S/A P
nush posterior S/A S
nun posterior S/A A
ek posterior (PoM) S/A S
As can be seen, the categories of the relative and reference clause participant are not equivalent. For an
even more complex system, see Zariquiey (2011) and this volume.
Relation between the relative and controller participant: the nature of identity.
One repeated observation made with respect to switch reference systems, even when the relevant
category is the syntactic subject for both the marked and the controller clause, is that there may be a
certain degree of flexibility when it comes to what counts as referential identity. We already saw
above that in Diyari in (5) above, subjects of reference clauses are considered identical to the reference
clause subject if the reference set of the former includes the one of the latter. Because the opposite
situation does not yield identity marking, this can be called an asymmetrical system (Comrie 1983).
Other languages may have a more symmetrical system (see also Davies & Comrie, this volume), such
as Huichol.
(28) Huichol [UTO-AZTECAN], Comrie 1983: 26-27
a. taame te-haataʔazɨa-ka nee ne-petɨa
we 1PL-arrive-IDENT I 1SG-leave
‘When we arrived, I left.’
b. nee ne-haataʔa-ka tanaitɨ te-pekɨɨ
I 1SG-arrive-IDENT together 1PL-leave
‘When I arrived, we left together.’
Further differences between languages in terms of what identity entails, as discussed by Davies &
Comrie, this volume, is that some systems seem to be sensitive only to whether or not the person value
changes, and are indifferent to differences in number value. Schmalz, this volume, discusses similar
facts for Yukaghir languages, emphasizing that interaction of (non-)inclusion of one participant set
into the other also works the other way as well, leading to “unexpected” non-identity marking. For
instance, a clause combination with two first person plural participants can still trigger different
subject marking if the sets of participants are different.
The markers
A number of variables relate to the morphological marking, or flagging of the SR construction. In the
first place, there are three theoretical possibilities in terms of markedness patterns, as indicated in
Figure 2, where a grey colored box represents a morphological marker of phonetic substance and the
white boxes represent no identifiable marker.
Identity Identity Identity
Non-identity Non-identity Non-identity
Figure 2: Major morphological markedness patterns
Apart from this general characterization, SR markers are subject to general variables of morphological
marking (Bickel & Nichols 2007):
Fusion refers to the degree to which a morpheme is integrated with its host. Values, from least to most
fused, are free, bound, and nonlinear. Most SR markers seem to be bound markers, though
phonologically free forms (especially in North American languages - see Jacobsen 1983) and
19
nonlinear markers (like nasalization of the final vowel in Jivaroan languages - see e.g. Overall 2007
for Aguaruna) also occur.
Flexivity indicates whether a morphological marker has lexically determined (item-based)
allomorphs. Values are flexive, and nonflexive. Flexive SR markers seem to be very rare, although it
does seem to occur in some Uto-Aztecan languages, like Cupeño, which has three phonologically
unrelated markers for different subject clauses, -qali, -weni, and -lee. The first two have different
functions (singular and plural subjects, respectively) but the third has a very restricted item-based
distribution, occurring only with the verb qal ‘be, dwell, sit in a place’ (Hill 2005: 408, see also Hill,
this volume).
(29) Cupeño [UTO-AZTECAN], Hill 2005: 410
puy-lya’a-ch-i te~tew-qa’ pe-qal-lee.
dine-INS-NPN-OC DUP~see-PRES 3S-be.there-NONID
‘I see the table where it was left.’
Exponence refers to the degree to which different categories are grouped together in a single
morpheme, with values cumulative or separative). Cumulative exponence is relatively common for SR
markers, for instance with person agreement (e.g. some Trans-New-Guinean languages, Jivaroan
lannguages), with gender and number (e.g. Tucanoan languages), with relative time marking (e.g.
Panoan languages), or with mood (e.g. Yurakaré - Van Gijn 2006), the latter type of cumulative
exponence is exemplified in (30), contrasting realis (30a), marked with -ja, with irrealis (30b) marked
with -ya.16
(30) Yurakaré [ISOLATE], Van Gijn 2014: 295 & 303
a. ti-bëjta-ø-ja ti-la-mala-ø samu
1SG-see-3=IDENT.REA 1SG-MAL-go.SG-3 jaguar
‘When the jaguar saw me, it ran away from me.’
b. tëshshu bëjta-ya bobo-ishti
weasel see-IDENT.IRR kill-FUT:1SG
‘If I see the weasel, I’ll kill it.’
Parameters relating to larger structures are position, referring to the relative position with respect to its
host of a morpheme (values: prae, post, in, simul). SR markers overwhelmingly more common follow
their hosts, though we have seen prefixes as well above for Lenakel. A final parameter refers to locus
of marking. This parameter is traditionally not applied to SR, but it could if one considers the
reference clause to be the head, and the marked clause the dependent. It seems to be a universal of SR
(cf. Bickel 2010) that it is marked on the dependent.
Apart from this, markers may differ in the host they attach to. In Haiman & Munro’s (1983b)
definition, SR markers are inflectional categories of the verb, but this is not necessarily so. SR markers
may be free forms (as in Maxakalí discussed above), or clitics that attach to larger strings than just the
verb, for instance the clause, or a group of words associated with the verb.
(31) Kiowa [KIOWA-TANOAN], McKenzie 2012: 56
a. John é zón cútā-dàu=chḕ é têm
John 3SG:3IN pull-out.PFV pencil-IN=WHEN.IDENT 3SG:3IN break.PFV
‘When John pulled it out, the pencil broke (in two).’
b. John cútā-dàu é zón=chḕ é têm
John pencil-IN 3SG:3IN pull-out.PFV=WHEN.IDENT 3SG:3IN break.PFV
‘When John pulled out the pencil, he broke it (in two).’
The identity marker =chḕ attaches to the last element of the clause, which is ‘pencil’ in (31a) because
of right-dislocation, or the clause-final verb in (31b)
16 In addition, the irrealis forms are not marked for subject, whereas the realis forms are.
20
Of course, one may argue that some of the variable values discussed in this section represent non-SR
forms, but we think in general that a more insightful picture can be obtained when all these different
possible configurations are regarded as typological variables. Only in this way can we hope to
understand the diachronic paths associated with the rise and/or decay of SR systems.
The next section discusses SR from different theoretical perspectives, focusing in particular on the
question what the function of SR is.
4. Theoretical perspectives on SR
Differences between theoretical approaches to SR mostly revolve around two related issues: the
function and the nature of SR. With regard to the first, there is a basic three-way opposition between
those who defend the idea that the function of SR is to track reference, a second strand of research
argues that SR is about event continuity, and a third approach advocates the fact that SR is part of a
family of constructions that relate to binding and control, and that they have no, or only secondarily,
pragmatic functionality. With respect to the second issue there are two opposite poles (as well as
intermediate positions): one that emphasizes the syntactic nature of SR, the other that defends the
pragmatic nature of SR.
There are some obvious correlations between the positions taken by authors on these two issues:
the advocates of switch reference as marking event continuity regard SR as pragmatic in nature, those
that see it as a specific instance of syntactic binding analyze SR as a syntactic phenomenon; in the
reference-tracking group of approaches, there is more variation with respect to the nature of SR. In the
next subsections we give concise overviews of the different theoretical positions. We start with SR as
a syntactic binding phenomenon (Section 4.1) to SR as a reference-tracking device (Section 4.2) and
finally to SR as a system for signaling event-continuity (Section 4.3), going from the most syntactic to
the most discourse-oriented approach.
4.1. SR as a syntactic mechanism
A radically syntactic approach to SR is taken by Finer (1984, 1985). In the first place, Finer argues
against SR having any pragmatic functionality. For SR to have a reference-tracking (disambiguating)
functionality, one would expect systems where only third persons are marked for same/different
reference, and in fact most systems include first and second persons as well, where from the point of
reference tracking SR has no functionality. He argues against a discourse cohesion (processing aid)
account where continuations or changes of topical participants are signaled, on the basis that one
would expect cataphoric SR marking only, forewarning the hearer that a topic shift or continuation is
to be expected in the next clause. Finally, Finer notes that pragmatic approaches to SR cannot explain
why SR does not occur in coordinate clauses (an observation that is not shared by everybody).
On the other hand, a number of further observations lead Finer to the suspicion that SR is a
syntactic phenomenon which, moreover, is connected to syntactic binding. He summarizes these
observations as follows (1985: 53):
(32) a. SS signals obligatory coreference between subject NPs of hierarchically adjacent clauses.
b. DS signals obligatory non-coreference between subject NPs of hierarchically adjacent
clauses.
c. The same-subject or different-subject relation is determined strictly locally.
d. Switch-reference involves subjects only.
The above four observations (which are all controversial) lead Finer to connect SR to Binding Theory
as proposed by Chomsky (1981: 188):17
(33) Binding Theory
17 An anaphor in Government & Binding and subsequent generative approaches to language refers to reflexives
and reciprocals; pronominals are pronouns which are not reflexive or reciprocal; an R-expression (referential
expression) is a full NP expression that is inherently referential. Governing category refers to the minimal
domain that contains the anaphor/pronominal/R-expression, its governor, and an accessible subject.
21
Principle A: An anaphor is bound in its governing category
Principle B: A pronominal is free in its governing category
Principle C: An R-expression is free everywhere
These three principles regard A-positions (potential theta positions) and regulate the (non-)coreference
behavior of reflexives, non-reflexive pronouns and NP expressions. Finer argues that SR is part of a
more generalized version of the Binding Theory, which also includes A’-positions (non-A-positions).
A’ Positions can also be bound by either A positions or by A’-positions, leading to four potential
binding relations.
(34) Binding relations in an extended binding theory
a. A bound/free with respect to A (the classic Binding Theory)
b. A bound/free with respect to A’ (Wh-movement)
c. A’ bound/free with respect to A (binding of e.g. Romance reflexive clitics)
d. A’ bound/free with respect to A’ (Cyclic Wh-movement)
SR markers, in Finer’s approach, are A’-positions that occupy the COMP position. Identity markers
are A’-anaphors, whereas non-identity markers are A’-pronominals. Because they are in COMP
position, their governing category includes the immediately superordinate verb and its subject,
meaning that the subject of the superordinate must be coreferential with the SS marker or non-
coreferential with the DS marker. Situation (c) applies when the immediately superordinate clause has
an SR marker that itself refers to a higher clause, so that a binding chain arises until its resolution in
the highest clause.
Later work that builds on Finer (1984, 1985), like Broadwell (1997), Watanabe (2000), Camacho
(2010), and Georgi (2012) offers refinements, incorporation into newer theoretical (sub)models of the
Minimalist Program (Chomsky 1995), and analyses that connect SR to even more general principles,
like agreement (Camacho 2010) and control (Borer 1989, Georgi 201218). Nevertheless, the core idea
of a chain of dependencies between clauses in an asymmetrical relation for identity clauses and the
absence of these dependencies for non-identity clauses remains intact.
An interesting deviation from this general line of thinking comes from Keine (2013). Keine’s
proposal starts from the same observation as Stirling (see Section 4.3), that there seem to be so many
cross-linguistic deviations from a general pattern of referential (non-)identity of subjects that we
should perhaps consider the possibility that referential continuity is not a central part of SR at all.
Keine proposes to regard SR as the instantiations of a coordination head. The crucial difference
between non-identity markers and identity markers in this approach is that the former involve vP (or
high) coordination (i.e. coordination of the lexical verb with its arguments, including the agentive
argument), whereas identity marking is underspecified for what type of equivalent elements it
conjoins, and can be used for VP (or low) coordination, where the external agent argument is not
involved in the coordination.19 In this way, identity marking is in principle compatible with a structure
of two clauses with distinct subjects, and non-identity with clauses with identical subjects. Unexpected
(i.e. from a reference-tracking perspective) non-identity marking may occur if the events in the clause
are considered distinct - correlating with the fact that vP conjunction in some languages involves two
structures each with their own event variable. Unexpected identity marking may occur with vP
coordination that involves a so-called defective v (i.e. a verb that cannot assign an agent role, like
unaccusative verbs). Whether and when unexpected (i.e. from a reference tracking perspective) occurs
is subject to cross-linguistic differences of how the difference between high and low coordination is
interpreted, as well as to how the morphosyntactic functions of identity and non-identity markers are
specified (see Keine 2013 for details). This proposal moves away from accounts based on binding, and
is in many ways closer to the event-based approach advocated by, among others, Stirling (1993) which
is discussed in Section 4.3.
18 In fact, Georgi (2012) analyzes identity constructions as situation b in Finer’s list: there is only one argument
there, but it is moved to the superordinate clause. 19 Keine also distinguishes coordination of TPs which, at least in the languages that he considers, are non-SR
coordination, which is in complementary distribution with SR marking, at least in the languages that he
considers.
22
4.2. SR as a reference tracking device
Haiman & Munro (1983b) regard switch reference as a syntactic phenomenon in the sense that it
canonically indicates (non-)coreference of the syntactic category of subject. With respect to the
function of SR, however, they regard SR as a more pragmatically oriented device “to avoid ambiguity
of reference” (Ibid. xi). On the basis of this functionality they predict that no language should exist
that marks SR for first and second, but not for third persons, since disambiguation for first and second
persons is redundant, but not for third persons. They do allow for languages that have SR systems that
are “generalized beyond the call of functional duty”. SR, in other words, is regarded as a reference
tracking device in Haiman & Munro’s view.
The perspective on the functionality of SR as a reference tracking mechanism is elaborated in more
detail by Foley & Van Valin (1984) who contrast switch-reference systems with switch-function
systems, in which “a particular participant is tracked across clauses, and the verbal morphology in
each clause signals the semantic function of that participant in that clause” (Ibid: 354).20 In a typical
switch-function system, reference tracking takes place by keeping a topical participant in subject
position, with voice morphology indicating the role of the tracked participant in the different events. In
contrast, switch-reference systems track a particular (syntactic or semantic) function, and signal
continuities or changes in the participant fulfilling that function. To illustrate the difference between
these tracking devices, consider (35):
(35) John rushed to work, Ø crossed the street in a hurry, and Ø was run over by a car.
In (35), the participant John is tracked across three events, the semantic role of John in the first two
events is agent, in the last event it is patient. The passive construction signals this latter fact, while
allowing for the tracked participant to remain in the privileged subject position.
A SR language would deal with (35) in quite a different way: they would simply track whether
there is a switch of participant for a particular function. Languages with a switch-reference system, in
the approach advocated by Foley & Van Valin may track different types of functions. The monitored
function may be invariably syntactic (subject) which corresponds to Haiman & Munro’s canonical SR,
but they may also be semantic in nature, as was discussed for Eastern Pomo above. The tracked
functions may furthermore be pragmatic in nature (or perhaps they are more accurately called
pragmatically influenced syntactic functions - see Matić et al. 2014), described as variable syntactic
pivots “in which the selection of the argument to function as pivot of a transitive verb is not
predictable from its semantic role and may be influenced by discourse-pragmatic considerations, in
particular the topicality and activation status of its referent” (Van Valin & LaPolla 1997: 291), like the
Barai case discussed in example (26) above.
A different, but compatible approach to SR is proposed by Kibrik (2011), who regards SR within a
broader context of reference in discourse, and general principles of human cognition. In his approach,
SR markers are part of a system of attending to a referent (reference), in which referential choice
depends on working memory. The more activated a referent in the discourse, the more likely it is
referred to by reduced referential devices such as pronouns, bound person markers). The process of
referential choice can be represented schematically as in Figure 3 (Ibid: 64).
Figure 3: The cognitive multi-factorial model of referential choice.
20 Foley & Van Valin (1984) distinguish further possible reference tracking systems (see Van Valin & LaPolla
1997 for a more recent and elaborate account), but they are not relevant to the topic of the present paper.
Discourse context
Referent’s
aggregate
activation Referent’s internal
properties
Referential
conflict filter Referential choice
Act
iva
tion
fact
ors
23
SR markers in this view are referential aids, which help distinguish between two or more activated
referents. In other words, their functionality lies in the referential conflict filter. Switch-reference, like
e.g. logophoricity, is considered to be a current sorting (i.e. it does not rely in any inherent property of
the referent) with a narrow domain (below the discourse level). Kibrik is aware that SR markers often
have functionalities that go beyond disambiguation, but he stresses that this “does not negate the
observation that switch-reference markers are often instrumental as a referent sorting” (Ibid: 341).
4.3. SR as a marker of event (dis-)continuity
As mentioned, in many languages, identity markers or non-identity markers pop up in “unexpected”
contexts, i.e. going counter their assumed referential interpretations. Some researchers have proposed
that the way to deal with these “discrepancies” is to consider the possibility that SR is primarily not
about referential continuity, but rather about broader principle.
Tsafiki was mentioned as a languages where “unexpected” SR marking may occur, see example (4)
here repeated as (36) for convenience:
(36) Tsafiki [BARBACOAN], Dickinson 2002: 137
junni man=ja-na-sa wata=te aman chide la-ri-bi
then again=come-PROG-NONID year=LOC now bone come.out-CAUS-PURP
man=ji-man-ti-e
again=go-SIT-REP-DECL
‘They say then, coming back, after one year he went to take out the bones.’
The fact that a non-identity marker appears, in spite of the identity of the two subjects is dues to the
fact that the events are temporally separated.
Another type of pragmatic influence on the SR system is presented by Amele, where “unexpected”
non-identity marking is determined by changes in the situation other than a referential switch. The
non-identity marker -co in (37) is claimed by Roberts (1988) to indicate a change of place.
(37) Amele [TRANS-NEW-GUINEA], Roberts 1988: 61
age ceta guldo-co-bil l-i bahim na tac-ein
3PL yam carry-NONID-3PL go-IDENT floor on fill-3PL-REM.PST
‘They carried the yams on their shoulders and went and filled up the yam store.’
In (38) from Central Pomo, the identity marker -hi is used in spite of distinct subject referents, due to
the fact that the events form a thematically coherent sequence.
(38) Central Pomo [POMOAN], Mithun 1993: 126
Ɂá· mkhe khčé=Ɂel dó-č-hi mí=·li ma Ɂdí-m=Ɂkhe
1A 2A bridge=the make-SML-IDENT that=with 2PAT take.PL-across=FUT
‘I will build the bridge for you and on that you'll take them (across)' = 'I will build you a
bridge to take them across on.’
A line of inquiry to deal with cases such as (36)-(38) is proposed by Stirling (1993), who argues for a
broader functionality of switch reference. In her perspective, switch reference is about congruence
between eventualities, of which reference tracking forms a sub-function. Nevertheless, she also notes
that referential (dis)continuity always forms a part of SR systems. She identifies six pivots of switch
reference systems (1993: 150-151):
- referential (dis)continuity: this includes the ‘classic’ or ‘canonical’ form of SR systems as
defined by Haiman & Munro, even though the data that Stirling discusses call for a more
liberal approach than the strict syntactic subject approach.
- agentivity value of important protagonist: unexpected DS marking may occur when the
reference remains constant, but the agentivity value of that participant changes. Unexpected
SS marking may occur when the syntactic subjects are not coreferent, but the subject of the
24
reference clause does not introduce a new agentive participant (e.g. in the case of impersonal
constructions).
- tense/time of the event: this pivot mainly concerns that the events agree in time, i.e. they take
place in what is considered to be the same moment or stretch of time.
- location of the event: in some languages, DS marking may occur under identity of subjects if
the two events take place at different locations
- mood of the clause. In some languages, there is a basic opposition between agreement
between realized or non-realized events which is marked by the switch reference system.
- continuance or shift out of a cohesive sequence of events: Stirling reports of languages
(Amele, Yankunytjatjara) where DS markers may be used to indicate an unexpected, surprise
change in the expected course of events.
In Stirling’s approach, identity is about agreement between aspects of eventualities, non-identity is
about disagreeing in at least one of the eventuality parameters.
Based on Stirling’s work Huang (2000) outlines a possibility that principles of SR systems can
possibly be reduced to more general, neo-Gricean principles. In this approach, cooperative
communicative behavior requires that continuity of eventualities be encoded by identity constructions.
If a non-identity encoding strategy is used, this leads to a Q-implicature (the continuity cannot be
maintained). Consistent with the cross-linguistic dominance of referential continuity in SR systems,
the preferred I-interpretation (the I Principle is the maxim of minimization) is a referential switch. If
this is not the case, the non-identity construction is taken to mark some other discontinuity between
eventualities, like time (as in Tsafiki) place (as in Amele). The precise interpretational scheme is
language specific (Ibid. 301).
McKenzie (2012) offers an alternative account of SR as (potentially) marking (shifts in) a topic
situation, which he defines as “a silent pronoun that refers to the part of the world that the sentence is
about, and on which its truth depends” (Ibid: 1). This silent pronoun may refers to the spatio-temporal
or thematic contexts against which a proposition is evaluated. In SR (sub)systems that are sensitive to
topic situations, identity clauses must contain two clauses whose propositions are evaluated against the
same topic situational context. In this way, scene-shifting effects occur, leading to “unexpected” non-
identity marking, and “unexpected” identity marking may occur when the topic situation for the two
propositions is identical in spite of the fact that there are different subjects. McKenzie furthermore
claims that topic-situation tracking takes preference over subject tracking (whenever a topic situation
can be tracked it will be tracked in a SR system). Since all matrix clauses have a referential topic
situation, and since coordination is the combining of two matrix clauses, it follows that SR in
coordinate clauses always tracks topic situations. In subordinate clauses, subjects are tracked.
5. Diachronic development of SR systems
The diachronic development of SR systems is still an understudied topic, and there is no clear
overview yet of the ways in which SR systems emerge and develop. In this section we discuss a
number of proposals that have been made with respect to emergence scenarios of SR systems (5.1),
and in Section 5.2, because contact-induced accounts of SR have been a recurring topic in SR-related
research, we give a concise characterization of a number of “SR areas” in the world.
5.1. Origins of SR
Haiman (1983), focusing on Papuan languages, discusses two types of SR systems: (i) systems where
the main difference between identity and non-identity medial clauses consists of the lack of (local)
person marking in identity clauses (reduction strategy - see Matić et al. 2014), and (ii) systems in
which the main difference between identity and non-identity medial clauses is that the latter is marked
by some additional morpheme (addition strategy - see Matić et al. 2014).
Haiman’s diachronic argument for the reductionist systems that he discusses is that they have their
origins in conjunction reduction strategies, where shared participants may be gapped. A system that
comes close to the purest form of the gapping scenario is from Ono, where non-identity verbs are
25
inflected for person (with different forms than found in reference clauses)21 and identity verbs carry no
person inflection.
(39) Ono [TRANS-NEW-GUINEA], Haiman 1983: 108, based on Wacke 1931
a. ngauk ne-ki ari-mai-ke
tobacco smoke-3SG.NONID go-PROG-3SG
‘Hei had a smoke and hej left.’
b. ngauk ne-ø ari-mai-ke
tobacco smoke-IDENT go-PROG-3SG
‘He had a smoke and left.’
Haiman mentions several different variations on this theme, which may involve additional marking for
both the identity and non-identity clause, but where the main difference is still that the identity clause
has no person marking (except in some instances anticipatory person marking which agrees with the
reference clause) whereas the non-identity clause does.
In languages of this type, Haiman argues, this situation applies in coordinate constructions22 only,
and the position of the ‘gap’ is dependent on whether the person markers are prefixes or not. He
connects this to the fact that in gapping, these two requirements also apply (coordination requirement
and the branching-dependent position of the gap), thus arguing for gapping as one of the constructions
that may give rise to a SR system.23
In languages of the second type, the non-identity clause differs from the identity clause in that it
contains a (usually invariable) additional marker that is absent in identity constructions. An example
of this type of language is Maring.
(40) Maring [TRANS-NEW-GUINEA], Haiman 1983: 115, based on Woodward 1973
a. pee-ba
go-3SG
‘He went and...’
b. pee-ba-k
go-3SG-NONID
‘He went and (another)...’
The diachronic scenario that may give rise to languages of this type is a conjunction marker or
nominalizing particle which in principle has open reference, but which specializes into marking non-
identity, in Haiman’s words (Ibid.: 107): “the portion of the meaning of a category that is unique to it”.
A slightly different though not unrelated diachronic perspective is offered by Givón (1983) who
insists on viewing SR from the wider functional angle of topic continuity, and also has a more liberal
interpretation of what constitutes a SR system than Haiman and also than the present chapter. He
reviews five coding strategies that can be connected to topic (dis)continuity cross-linguistically, given
in Table 6 (Ibid. 69):
Table 6: Topic (dis)continuity coding strategies with grammaticalization potential (Givón 1983) Continuity coding devices Discontinuity coding devices
unstressed / bound / zero pronouns stressed / independent pronouns
non-finite/participial/nominalized clauses finite or other subject-marked clauses
comment-topic order topic-comment order
deictic proximity deictic remoteness
21 Haiman does not specify the origins of these medial person markers, but suggests that there may be a variety
of origins in different languages. 22 Haiman defines coordination on the basis of the number of features the two subclauses have in common: the
more features they share the more coordinate the construction. This differs from the conception of coordination
in e.g. Finer (1984, 1985) discussed above. 23 Note that, in our approach to SR, pure reduction strategies, where the only difference between identity and
non-identity is the lack of person expression in the former, are not considered canonical SR systems since they
do not involve any additional flagging. Ono qualifies as an SR language because the person markers of the non-
identity clauses differ from those in an independent clause.
26
pronominal first person pronominal third person
Givón argues that each of the five coding contrasts in Table 6 has a certain degree of potential to
develop into more grammaticalized systems such as SR. The top coding strategy, unstressed/zero
versus stressed pronouns, can be connected to Haiman’s first scenario. An example of a
grammaticalized system on this basis given by Givón is from Lango, on the basis of a personal
comment by Michael Noonan.24
(41) Lango [NILOTIC], Givón 1983: 70
a. lócà ò-támó ní ò-nęnò gwók
man he-think that he-saw dog
‘The mani thought hei saw the dog.’
b. lócà ò-támó ní ę-nęnò gwók
man he-think that he-saw dog
‘The mani thought hej saw the dog.’
The difference in form between the two types of third person markers can be traced back to the bound
versus free pronouns of Lango, the former having developed into identity markers, and the latter in to
non-identity markers.
The second coding principle is not unlike the first in that finite structures. or structures in which the
subject participant is marked in some way, are more informative than non-finite structures and the
principle of economy will predict that you will find the less informative structures in situations where
this does not lead to interpretative difficulties. In addition, as was also pointed out by Haiman, the
specific markers of dependency, be they nominalizers or conjunction markers, can develop into
identity and non-identity markers.
The bottom three contrasts are, in Givón’s view, less likely to give rise to grammaticalized
reference-tracking systems. Word order differences are contrastive as markers of relative prominence,
but not as contrastive as the more powerful zero versus non-zero contrast, which form the more
extreme points on the continuum of treatment of topic and comment structures, and are therefore more
likely to grammaticalize than the word order contrasts, see Figure 4 (Givón 1983: 72)
predictability of topic
+ 1
zero topic > comment-topic order > topic-comment order > zero comment
Figure 4: Grammatical treatment of topic and comment
The proximate versus distal oppositions in deictic expressions (corresponding to given/definite
versus new/indefinite information) is not expected to give rise to systems where (dis)continuity of
subjects is at stake, since subjects tend to be overwhelmingly definite. Nevertheless, Givón gives an
example from Persian, where topic continuity marking (which, in Givón’s view, belongs to the same
functional domain as SR marking) is based on the distal versus proximate distinction.
Givón’s final coding strategy is restricted to the Tibetan language Sherpa, where subject agreement
(limited to perfect aspect) contrasts speaker versus non-speaker. When these perfect structures form
complements of verba dicendi, they may give rise to SR-like systems (on the basis of a personal
communication by Konchcchok J. Lama).
(42) Sherpa [SINO-TIBETAN], Givón 1983: 76
a. ti-gi si-kyaa-sung cenyi caax-yin
he-ERG say-AUX-PERF.2/3 cup break-PERF.1
‘He said: “I broke the cup”’ >> ‘Hei said hei broke the cup.’
b. ti-gi si-kyaa-sung cenyi caax-sung
he-ERG say-AUX-PERF.2/3 cup break-PERF.2/3
24 Givón classifies the Lango system as a SR system, but it is rather a logophoric system. Nevertheless, the
general point remains that the opposition mentioned by Givón can lead to a more grammaticalized reference
tracking device.
27
‘He said: “He broke the cup”’ >> ‘Hei said hej broke the cup.’
A further often mentioned source for SR markers are case markers, in particular with respect to
contact-induced diffusion of SR systems, through a process of exaptation. The first to draw attention
to this in a systematic way was Austin (1980, 1981). Austin (1981) discusses various Australian
languages with a switch-reference system and argues for an account of the distribution of SR
languages in terms of contact-induced diffusion. One of his arguments is that areal patterns can be
discerned on the basis of correspondences between nominal case markers and SR markers in relative
clauses. Three contiguous geographical areas can be distinguished: one where there is a
correspondence between locative case and non-identity markers, one where there is a correspondence
between locative case and identity markers, and a third where there is a correspondence both between
locative case marking and identity marking and between allative case and non-identity markers. The
connection between case markers and SR markers has been reported for other, unrelated languages or
language families, like the Tacanan (e.g. Guillaume 2011, Vuillermet 2014), Panoan (e.g. Valenzuela
2003, Zariquiey 2011) and Jivaroan (Overall 2007) families in South America, Muskogean (but see
McKenzie 2012 for a critique), Yuman, Uto-Aztecan families in North America (Jacobsen 1983). An
important issue in the potential paths of emergence of SR systems is the type of clauses for which a
SR system grammaticalizes. As mentioned above, the gapping source was considered by Haiman to be
indicative of a grammaticalization path through coordinate-like clauses. Austin emphasizes the role of
(nominalized locational) relative clauses in the development of the locative cases into SR markers,
while Jacobsen regards case marker origins to point to paratactic structures. It is clear that much more
work is required to uncover the role of clause types in possible diachronic paths of SR systems.
McKenzie (2012), although he rejects any direct connections between SR systems and other
systems in languages, draws attention to the importance of the process of exaptation to the study of
SR, which he describes as “the co-opting of morphemes from one module of grammar to another,
without changing phonetic form, in order to serve a different function” (Ibid. 71). For McKenzie,
exaptation is often triggered by language contact, and a resulting pattern replication (i.e. the borrowing
of abstract patterns or subsystems without the transfer of any phonetic substance from one language to
another). This scenario still leaves some questions unanswered, in particular why pattern replication is
more common for SR systems than morpheme borrowing, and why certain morphological sources of
SR markers are more common than others.
5.2. SR areas
As mentioned above, SR tends to occur in geographical clusters, and often across genealogical
boundaries. It is therefore not surprising that people have regarded the distribution of SR systems to be
influenced heavily by language contact. We briefly characterize a few of these “SR areas” in this
section.
Western North America
According to the survey in McKenzie (forthc.), 68 North American languages have an SR system,
spread over 12 families and 4 isolates. The presence of SR systems is clearly geographically skewed,
with the bulk of the SR systems found in a contiguous area in the (south) west of the United States and
adjacent areas in north-western Mexico. Compared to other SR areas, SR in North American
languages is found in a relatively large range of clause types: coordinate, relative, and complement
clauses, as well as in semantically underspecified clause chains. Jacobsen (1983) translates this into a
division into two types of languages, those which have SR only in adverbial clauses, and those that
have SR in adverbial clauses as well as in other clause types, which could be translated in the form of
an implication: if a NA language has SR in non-adverbial clauses, it has SR in adverbial clauses.
McKenzie (forthc.) proposes another implication for North American SR: although a number of
reports on individual North American languages have revealed that there are quite some deviations
from the “canonical” subject pivot, in favour of marking inter-event coherence. Nevertheless,
McKenzie observes that canonical SR is always part of the function of SR markers in North American
languages, yielding the implication that if a language has non-canonical SR, it also has canonical SR
(in McKenzie’s approach meaning tracking subject reference). In terms of their morphology, most SR
systems in North America have two opposed values (identity and non-identity) and the marker
28
function independently from agreement paradigms. None of the SR markers is a proclitic or prefix, but
SR markers can be suffixes, enclitics, or root forms, generally “found near the edges of the pivot
clause [i.e. the marked clause - RG & JH], though not exclusively” (McKenzie forthc. [21]).. Also, in
a number of cases the identity and non-identity markers are asymmetrical in terms of their
morphological categories.
Jacobsen (1983) is careful in drawing any conclusions about diffusion as a possible reason for the
distribution of SR systems in North America, saying that “the centrally located Yuman and nearby
Uto-Aztecan languages (Hopi and Papago) may be a prime source of this influence [i.e. contact-
induced diffusion of SR - RG & JH]”, based on the fact that as one moves away from this central area,
SR is less prominently present in the languages.
Western South America
SR systems are particularly common in Western South America, both in the Andean mountain range
(Quechuan, Uru, Chipaya, formerly also Aymaran) and the adjacent upper Amazon area (e.g. Tacanan,
Panoan, Barbacoan, Tucanoan, Jivaroan). SR has also been reported for Eastern Macro-Jê languages
(Rodrigues 1999) and a SR-like system is found in a number of Tupian languages, which are spread
over the entire northern half of the continent.
The SR systems found in western South America have a number of traits in common. First, in
almost all cases, both identity and non-identity situations are morphologically marked, almost
invariably by a suffix, in some cases by an enclitic. The most common clause type where SR is found
is a (co-)subordinate adverbial clause, usually with temporal meaning; other common interclausal
semantics are conditional, reason, concession, and purpose. In a few language families (Panoan,
Jivaroan) longer clause chains are common, and in most SR languages tail-head linkage seems to be
frequent. In terms of the controller and target categories, most languages have syntactically defined
subject pivots, but Panoan languages in particular allow for different types of coreference pivots,
involving intransitive subjects (S), transitive subjects (A), and objects (P). An understudied part of
South-American SR is its usage patterns. For a few languages, a number of factors have been
identified that interfere or override the automatic application of SR based on the pivot categories, to do
with accessibility, focus, referentiality, and event coherence.
In spite of the similarities between some of the systems in South America, the SR systems in most
families show coherent family-internal patterns, with peculiarities that are specific to SR in that
language family, e.g. with respect to cumulative exponence of the SR markers, controller and target
categories, and interclausal semantics. This suggests that SR has been a part of those families at the
time the languages started to diverge. For more information, see the contributions by Van Gijn and
Floyd & Norcliffe in this volume.
West and central Australia
Austin (1981) reports the occurrence of SR in a number of unrelated languages in a large continuous
area in central and south-western Australia. Austin surveys 6 language groups with SR systems:
Arabana-Wangganguru, Nyungic, (Western Desert, Mantharta, Kanyara, Ngarrka), Arandic, Wagaya,
Garawa-Wanyi, and Djingili. The Nyungic and Mantharta groups belong to the Pama-Nyungan
family.
Australian SR systems display a good deal of overlap. SR occurs between subordinate and main
clauses, and is marked by a suffix on the subordinate verb. The pivot is defined syntactically as the
subject (S and A) both for the controller and the target, and most systems work with a binary contrast
of identity versus non-identity, although there are exceptions to this pattern, for instance in systems
where the number of markers is increased by the fact that the identity marker also expresses the
syntactic function of the coreferential main clause NP (Warlpiri and Warlmanpa). SR marking is found
in relative clauses (with a sometimes wide range of semantics) and in purpose clauses. Perhaps most
strikingly from a contact perspective, the distribution of clause types with SR follows a clear areal and
non-genealogical pattern. The languages to the north and south of the SR area display SR in relative
clauses only, while the ones in the middle and to the west tend to have SR in both relative and purpose
clauses. The other areal phenomenon, mentioned above, is the source of some of the SR markers from
spatial case markers. If these facts are coupled with the fact that the phonetic form of the markers does
not suggest direct borrowing, the situation in Australia, at least for some of the markers, seems to be
most elegantly explained by contact-induced exaptation as discussed above.
29
Papua New Guinea (and Oceania)
According to Roberts (1997: 102) SR “occurs in more languages that are geographically adjacent than
anywhere else in the world”. In his sample of 169 languages, Roberts (Ibid. 122) reports that 122
(70%) have a SR system. The SR languages of Papua New Guinea are by and large spoken in the large
geographically contiguous highland area covering the central, north-eastern and south-eastern part of
the mainland, as well as some non-contiguous areas particularly on Bougainville Island. SR extends to
a few Austronesian languages spoken on islands to the northeast of PNG as well. Further removed, the
Austronesian languages spoken on Vanuatu display a (rather different type of) SR. The area may be
larger, but Roberts only surveys PNG. Because of ongoing discussions about the genealogical
relatedness between the different PNG languages, it is difficult to discern the genealogical and areal
contributions to the current situation.
Papuan SR structures are characterized by potentially large strings of medial verbs and a finite
anchor clause. In most languages these medial verbs are semantically rather vague or unspecified and
their interpretation derives from context. There are also a handful of languages that have more than
one type of medial verb depending on the interclausal semantics. There is quite some variation in
terms of the formal encoding of SR between the languages. Roberts (1997: 136) distinguishes seven
strategies, the most common of which is a morphological opposition between two dedicated
morphemes, another common way to mark non-identity is by marking its subject (often with a special
set of person markers). A special feature of the Vanuatu languages is the marking of the anticipatory
subject, as briefly discussed above under the heading of echo subjects. Finally, an important feature of
some PNG languages is that their SR systems do not track the syntactic subject category, but rather the
semantic category of agent (like Alamblak), topicality (like Barai) or spatio-temporal cohesion
(Amele).
Ethiopia
Africa does not belong to the traditionally recognized SR areas and is highly understudied in this
respect. A recent paper by Yvonne Treis (2012) argues for a bona fide SR area in south-western
Ethiopia, where a number of East Cushitic languages have developed SR systems under the influence
of North Omotic languages.
In the Cushitic languages Kambaata, Alaaba, Hadiyya, and Sidaama, SR is marked for a subset of
the available converbs - verb forms that are reduced in inflectional potential used to encode mostly
adverbial functions. The SR converbs generally precede the controller clause and they may express a
variety of semantic relations to the controller clause. The Cushitic languages differ in the extent to
which SR applies and the way in which it is marked. In Kambaata, SR seems to be most
grammaticalized. Subject switches are obligatorily marked on perfective and imperfective contexts,
and follow a rather strictly syntactic pivot-controller relation.25 Identity clauses are unmarked. A
similar situation holds for closely related Alaaba, except that DS marking is applied to only one
converb type. In Hadiyya different converbs are used for identity and non-identity contexts. The
system in Sidaama, finally, is similar to the one found in Kambaata, but non-identity is marked with an
enclitic =nna, a grammaticalized conjunction marker.
SR marking in the North Omotic languages is widespread. Like in the four Cushitic languages
mentioned above, SR in North Omotic languages is marked for (part of) the converbs. Different
converbial non-finite endings are attached to roots to distinguish identity and non-identity contexts.
Systems diverge from relatively simple two-way oppositions in Zargulla26 to more complex systems in
Yemsa where cumulative exponence of SR marking and interclausal relational semantics and person
marking leads to 14 different markers in the system.
The more widespread presence of SR in North Omotic languages and the geographical proximity
of the four East Cushitic languages to Omotic languages (in particular Yemsa and Wolaitta) suggest a
contact-induced change in the East Cushitic languages under Omotic influence, although the exact
mechanism involved remains unclear.
25 There is a set of SR marked purposive verb forms as well, which differ from the other converbs in that they are
not marked for (im)perfectivity (see Treis 2010 for more information). 26 There may be more oppositions in this language, but Treis can only reliably report two.
30
North-eastern Eurasia
To our knowledge, there is no overview of SR structures in Northeastern Eurasia, so the overview that
we can give here is particularly incomplete. Judging from the contributions of Schmalz (on Yukaghir
languages) and Matić (on the Tungusic language Even) SR in these languages is not unsimilar to what
Treis describes for Ethiopia. Like in SR-marking Cushitic and Northern Omotic languages, Yukaghir
and Even have a set of converbs that are marked for the type of semantic relation they have to the
main clause, as well as for SR. A difference with the African systems is that the Siberian systems
described in this volume seem to be more sensitive to semantic and/or pragmatic considerations. The
Even converbs, moreover, are nominalized constructions. It is unclear what the geographical extent of
this area is, but it is potentially enormous, as converbs in eastern Eurasia are common in many
different families (Mongolic, Tungusic, Yukaghir, Nivkh, Chukot, Korean, Turkic)27, although it is not
known - at least to us - to what extent these different converbal systems can be described in terms of
SR systems (e.g. see the contribution by Jendraschek in this volume against SR-sensitive converbs in
Korean and Turkish). Hopefully a (team of) areal specialist(s) will take up the challenge to describe
the converbs of these families so that we achieve a clearer picture of the extent and internal cohesion
of this potential SR area.
27 We are grateful to Dejan Matić for pointing this fact out to us.
31
Abbreviations
A agent
ABS absolutive
ASC associative case
ASP aspect
AUX auxiliary
CAUS causative
COM comitative
COMP complementizer
CONV converb
COREF coreferent
DECL declarative
DEF definite
DEM demonstrative
DL dual
DUP reduplicated form
ERG ergative
ES echo subject
EVID evidential
EXCL exclusive
FUT future
GER gerund
IDENT identity
IMPL implicational
IN inverse number
INCL inclusive
IND indicative
INS instrument
INTR intransitive
IRR irrealis
LDR long-distance reflexive
LOC locative
LOG logophoric
MAL malefactive
MED medial
NOM nominative
NONID non-identity
NONFUT non-future
NPN non-possessed noun
OBJ object
OBV obviative
OC object case
PAT patient
PERF perfect
PFV perfective
PL plural
PRES present
PROG progressive
PST past
PURP purposive
REM remote
REA realis
REP reportative
SG singular
SIT situational
SML semelfactive
SUBJ subject
U undergoer
References
Andrews, Avery. 2007. Relative clauses. In: Timothy Shopen (ed.): Language typology and syntactic
description. Volume II: complex constructions. Cambridge: Cambridge University Press, pp. 206-
236.
Austin, Peter. 1980. Switch-reference in Australian languages. In: Pamela Munro (ed.). Studies of
switch-reference. Los Angeles: University of California, pp. 7-47.
Austin, Peter. 1981. Switch-reference in Australia. Language 57 (2): 309-334.
Bickel, Balthasar. 2010. Towards a multivariate typology of reference tracking. Presentation given at
the Meeting of the Research Group 742 “Grammar and Processing of Verbal Arguments”.
Powerpoint available at http://www.comparativelinguistics.uzh.ch/bickel/presentations_en.html
Bickel, Balthasar & Johanna Nichols. 2007. Inflectional morphology. In: Timothy Shopen (ed.):
Language typology and syntactic description. Volume III: grammatical categories and the lexicon.
Cambridge: Cambridge University Press, pp. 169-240.
Borer, Hagit. 1989. Anaphoric AGR. In: O. Jaeggli & K. Safir (eds) The Null Subject Parameter.
Dordrecht: Kluwer,pp.69–109.
Broadwell, George Aaron (1997), Binding Theory and Switch Reference, in H.Bennis, P.Picaand & J.
Rooryck (eds.) Atomism and Binding. Dordrecht: Foris, pp.31-49.
Campbell, Lyle. 1997. American Indian Languages: the historical linguistics of Native America.
Oxford: Oxford University Press.
Camacho, José. 2010. On case concord: the syntax of switch-reference clauses. Natural Language and
Linguistic Theory 28: 239-274.
Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: FORIS Publications.
32
Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: The MIT Press.
Cole, Peter 1983. Switch-reference in two Quechua languages. In: John Haiman & Pamela Munro
(eds.) Switch reference and Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp.
1-16.
Cole, Peter, Gabriella Hermon, and C.T. James Huang. 2001. Long-distance reflexives: The state of
the art. In: Peter Cole, Gabriella Hermon, and C.T. James Huang (eds.): Syntax and Semantics 33:
Long-distance reflexives. San Diego: Academic Press, pp. xiii-xlvii.
Comrie, Bernard. 1983. Switch-reference in Huichol: a typological study. In: John Haiman & Pamela
Munro (eds.) Switch reference and Universal Grammar. Amsterdam & Philadelphia: John
Benjamins, pp. 17-38.
Corbett, Greville G. 2006. Agreement. Cambridge: Cambridge University Press.
Cristofaro, Sonia. 2003. Subordination. Oxford: Oxford University Press.
Crowley, Terry. 1998. An Erromangan (Sye) grammar. Oceanic linguistics special publication.
Honolulu: University of Hawaii Press.
Dickinson, Connie. 2002. Complex predicates in Tsafiki. PhD thesis University of Oregon.
Dixon, R. M. W. 1979. Ergativity. Language, 55 (1): 59-138.
Ferreira, Rogério Vicente. 2005. Língua Matis (Pano): Uma Descrição Grammatical. PhD. Thesis
Universidade Estadual de Campinas.
Finer, Daniel. 1984. The formal grammar of switch-reference. PhD thesis, University of
Massachusetts-Amherst, Amherst, Mass
Finer,Daniel. 1985. The syntax of switch-reference. Linguistic Inquiry 16,35-55.
Foley, William A & Robert D. Van Valin. 1984. Functional Syntax and Universal Grammar.
Cambridge/New York: Cambridge University Press.
Gabas Jr., Nilson,. 1999. A Grammar of Karo, Tupi (Brazil). Santa Barbara: University of California.
Ph.D. Thesis University of California at Santa Barbara.
Georgi, Doreen. 2012. Switch-reference by movement. In: P. Weisser (ed.) Perspectives on switch-
reference: local modeling and empirical distribution. Leipzig: Institut für Linguistik, pp. 1-40.
Gijn, Rik van. 2006. A grammar of Yurakaré. PhD. Thesis Radboud University Nijmegen.
Gijn, Rik van. 2014. Repeated dependent clauses in Yurakaré. In: Rik van Gijn, Dejan Matić, Jeremy
Hammond, Saskia van Putten and Ana Vilacy Galucio (eds.) Information structure and reference
tracking in complex sentences. John Benjamins, pp. 291-308.
Gijn, Rik van & Harald Hammarström. forthc. A construction‐based approach to measuring distances
between languages: subordination strategies in South America.
Givón, T. 1983. Topic continuity in discourse: the functional domain of switch-reference. In: John
Haiman & Pamela Munro (eds.) Switch reference and Universal Grammar. Amsterdam &
Philadelphia: John Benjamins, pp. 51-82.
Gordon, Lynn. 1983. Switch-reference, clause order, and interclausal relationships in Maricopa. In:
John Haiman & Pamela Munro (eds.) Switch reference and Universal Grammar. Amsterdam &
Philadelphia: John Benjamins, pp. 83-104.
Guillaume, Antoine. 2011. Subordinate clauses, switch-reference, and tail-head linkage in Cavineña
narratives. In: Rik van Gijn, Katharina Haude & Pieter Muysken (eds.): Subordination in Native
South American languages. Amsterdam & Philadelphia: John Benjamins, pp. 109-140.
Haiman, John. 1979. Hua: A Papuan Language of New Guinea. In: Timothy Shopen (ed.), Languages
and their speakers, 35-89. Cambridge: Winthrop.
Haiman, John. 1983. On some origins of switch-reference marking. In: John Haiman & Pamela Munro
(eds.) Switch reference and Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp.
105-128.
Haiman, John & Pamela Munro (eds.) 1983a. Switch reference and Universal Grammar. Amsterdam
& Philadelphia: John Benjamins.
Haiman & Munro 1983b. Introduction. In: John Haiman & Pamela Munro (eds.) Switch reference and
Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp. ix-xv.
Hammond, Jeremy. 2014. Switch-reference antecedence and subordination in Whitesands (Oceanic).
In Rik van Gijn, Jeremy Hammond, Dejan Matic, Saskia van Putten & Ana Vilacy Galucio (eds.),
Information Structure and Complex Sentences. Amsterdam: John Benjamins, pp. 263-290.
33
Haspelmath, Martin. 1995. The converb as a cross-linguistically valid category. In Martin Haspelmath
& Ekkehard König (Eds.): Converbs in Cross-Linguistic Perspective. Berlin, New York: Mouton
de Gruyter, pp. 1-55.
Haspelmath, Martin. 2007. Coordination. In: Timothy Shopen (ed.): Language typology and syntactic
description. Volume II: complex constructions. Cambridge: Cambridge University Press, pp. 1-51.
Headland, Paul & Stephen Levinsohn. 1977. Prominence and cohesion in Tunebo discourse. In: R.
Longacre & F. Woods (eds.): Discourse grammar Studies in the languages of Colombia, Panama,
and Ecuador, volume 2. Norman: Summer Institute of Linguistics, pp. 133-157.
Heath, Jeffrey. 1983. Referential tracking in Nunggubuyu. In: John Haiman & Pamela Munro (eds.)
Switch reference and Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp. 129-
150.
Hill, Jane A. 2005. A Grammar of Cupeño. Berkeley and Los Angeles: University of California Press.
Huang, Yan. 2000. Anaphora. Oxford: Oxford University Press.
Huang, C.-T. James & C.-C. Jane Tang. 1991. The local nature of the long-distance reflexives in
Chinese. In: J. Koster & E. Reuland (eds.) Long-distance anaphora. Cambridge: Cambridge
University Press, pp. 263-282.
Huisman, Ronald D. 1973. Angaataha verb morphology. Linguistics 110: 43-54.
Jacobsen, William Jr. 1967. Switch-reference in Hokan-Coahuiltecan. In: Dell Hymes and W. Bittle
(eds.): Studies in Southwestern Ethnolinguistics: Meaning and History in the Languages of the
American Southwest The Hague: Mouton, pp. 238-63.
Jacobsen, William H. 1983. Typological and genetic notes on switch-reference systems in North
American Indian languages. In: John Haiman & Pamela Munro (eds.) Switch reference and
Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp. 151-184.
Jensen, Cheryl. 1997. Coreferential marking in Tupí-Guaraní languages. Paper presented at the XIII
International Conference of Historical Linguistics, Düsseldorf, Germany August 10-15, 1997.
Handout available at: http://www-01.sil.org/americas/brasil/publcns/ling/ichlcor.pdf
Kalectaca, Milo. 1978. Lessons in Hopi. Tucson: University of Arizona Press.
Keenan, Edward & Bernard Comrie. 1977. Noun Phrase accessibility and universal grammar.
Linguistic Inquiry 8(1): 63-99.
Keine, Stefan. 2013. Deconstructing switch-reference. Natural Language and Linguistic Theory 31
(3): 767-826.
Kibrik, Andrej A. 2011. Reference in Discourse. Oxford: Oxford University Press.
Langdon, Margaret and Pamela Munro. 1979. Subject and (Switch-)Reference in Yuman. Folia
Linguistica 13.
Lee, Hyeran. 2003. Dependency of Long-Distance Reflexives. In: Dong Hong Ji & Kim Teng (eds.):
Proceedings of the 17th Pacific Asia Conference on Language, Information and Computation.
Sentosa, Singapore: COLIPS Publications, pp. 431–440. Available online at
http://aclweb.org/anthology/Y03-1048.
Lehmann, Christian. 1988. Towards a typology of clause linkage. In: J. Haiman & S.A. Thompson
(eds.) Clause combining in grammar and discourse. Amsterdam/Philadelphia: John Benjamins, pp.
181-226.
Longacre, Robert E. 1972. Hierarchy and universality of discourse constituents in New Guinea
languages. Washington: Georgetown University Press.
Longacre, Robert E. 1983. Switch-reference systems in two distinct linguistic areas: Wojokeso (Papua
New Guinea) and Guanano (Northern South America). In: John Haiman & Pamela Munro (eds.)
Switch reference and Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp. 185-
208.
Longacre Robert E. 2007. Sentences as combinations of clauses. In: Timothy Shopen (ed.): Language
typology and syntactic description. Volume II: complex constructions. Cambridge: Cambridge
University Press, pp. 372-420.
Lynch, John. 1983. Switch-reference in Lenakel. In: John Haiman & Pamela Munro (eds.) Switch
reference and Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp. 209-222.
Malchukov, Andrej. 2006. Constraining nominalization: form/function competition. Linguistics 44 (5):
973-1009.
Matic, Dejan, Rik van Gijn & Robert van Valin. 2014. Information structure and reference tracking in
complex sentences: an overview. In: Rik van Gijn, Jeremy Hammond, Dejan Matic, Saskia van
34
Putten & Ana Vilacy Galucio (eds.) Information structure and reference tracking in complex
sentences. Amsterdam/Philadelphia: John Benjamins, pp. 1-42.
McKenzie, Andrew. 2012. The Role of Contextual Restriction in Reference Tracking, PhD Thesis
University of Massachuesetts, Amherst.
McKenzie, Andrew. in prep. On certain switch-reference strategies. Manuscript.
McLendon, Sally. 1978. Ergativity, case, and transitivity in Eastern Pomo. International Journal of
American Linguistics 44(1): 1-9.
Mithun, Marianne 1993. ““Switch-Reference”: Clause Combining in Central Pomo”. International
Journal of American Linguistics 59.2: 119–136
Mithun, Marianne. 1999. The languages of native North America. Cambridge: Cambridge University
Press.
Munro, Pamela. 1976. Subject copying, auxiliarization, and predicate raising: the Mojave evidence.
International Journal of American Linguistics 42: 99-112.
Munro, Pamela (ed.) 1980a. Studies of switch-reference. Los Angeles: University of California Los
Angeles [UCLA Papers in Syntax 8].
Munro, Pamela. 1980b. Introduction. In: Pamela Munro (ed.). Studies of switch-reference. Los
Angeles: University of California, pp. 1-3.
Munro, Pamela. 1980c. On the syntactic status of switch-reference clauses: the special case of Mojave
comitatives. In: Pamela Munro (ed.). Studies of switch-reference. Los Angeles: University of
California, pp. 145-160.
Munro, Pamela. 1983. When “same” is “not different”. In: John Haiman & Pamela Munro (eds.)
Switch reference and Universal Grammar. Amsterdam & Philadelphia: John Benjamins, pp. 223-
244.
Nedjalkov 1995. Some typological parameters of converbs. In Martin Haspelmath & Ekkehard König
(Eds.): Converbs in Cross-Linguistic Perspective. Berlin, New York: Mouton de Gruyter, pp. 97-
136.
Nichols, Johanna. 1983. Switch-reference in the Northeast Caucasus. In: John Haiman & Pamela
Munro (eds.) Switch reference and Universal Grammar. Amsterdam & Philadelphia: John
Benjamins, pp. 245-266.
Noonan, Michael. 2007. Complementation. In: Timothy Shopen (ed.): Language typology and
syntactic description. Volume II: complex constructions. Cambridge: Cambridge University Press,
pp. 52-150.
O’Connor, Mary. 1993. Disjoint reference and pragmatic inference: anaphora and switch-reference in
Northern Pomo. In: W. Foley (ed.) The role of theory in language description. Berlin: Mouton de
Gruyter, pp. 215-242.
Olson, Michael. 1978. Switch reference in Barai. Proceedings of the fourth Annual Meeting of the
Berkeley Linguistics Society 4: 140-156.
Olson, Michael. 1981. Barai clause junctures: toward a functional theory of interclausal relations.
Australian National University. Canberra: Australian National University.
Oswalt, R. L. 1976. Switch reference in Maiduan: An areal and typological contribution. International
Journal of American Linguistics 42(4): 297-304.
Overall, Simon. 2007. A Grammar of Aguaruna. PhD dissertation LaTrobe University.
Payne, Doris L. 1980. Switch-reference in Chickasaw. In: Pamela Munro (ed.). Studies of switch-
reference. Los Angeles: University of California, pp. 89-118.
Payne, Thomas E. 1980. Who’s kissing who: the fourth person in Yup’ik Eskimo. In: Pamela Munro
(ed.). Studies of switch-reference. Los Angeles: University of California, pp. 65-88.
Popovich, A. Harald. 1986. The nominal reference system of Maxakalí. In: Ursula Wiesemann (ed.)
Pronominal systems. Tübingen: Gunther Narr Verlag, pp. 351-358.
Reesink, Ger. 1983. Switch reference and topicality hierarchies. Studies in Language 7(2): 215-246.
Reesink, Ger. 2014. Topic management and clause combination in the Papuan language Usan. In: Rik
van Gijn, Jeremy Hammond, Dejan Matic, Saskia van Putten & Ana Vilacy Galucio (eds.)
Information structure and reference tracking in complex sentences. Amsterdam/Philadelphia: John
Benjamins, pp.231-262.
Roberts, John R. 1988. “Amele Switch-Reference and the Theory of Grammar”. Linguistic Inquiry
19.1: 45–63.
35
Roberts, John R. 1997. ‘Switch-reference in Papua New Guinea: A preliminary survey.’ In A. Pawley,
ed.: Papers in Papuan linguistics, No. 3, 101–241.Pacific Linguistics, A-87.
Rodrigues, Aryon D. 1999. Macro-Jê. In: R.M.W. Dixon & Alexandra Y. Aikhenvald (eds.) The
Amazonian languages. Cambridge: Cambridge University Press, pp. 164-206.
Rudnev. 2008. Some syntax and semantics of long distance reflexives in Turkish and elsewhere.
Online publication available at
http://people.umass.edu/partee/RGGU_2008/TypologyReports/Rudnev_2008_Turkish_anaph_Fina
l.pdf
Scott, Graham 1973. Higher levels of Fore grammar (Pacific Linguistics B-47). Canberra: Australian
National University.
Shopen, Timothy (ed.) 2007. Language typology and syntactic description. Volume II: complex
constructions. Second edition. Cambridge: Cambridge University Press
Siewierska, Anna. 1999. From anaphoric pronoun to grammatical agreement marker: why objects
don’t make it. Folia Linguistica 33(2): 225-251.
Steward, Anne M. 1988. Switch-reference in Conchucos Quechua. In: William Shipley (ed.): In honor
of Mary Haas: From the Haas Festival Conference on Native American Linguistics. Berlin:
Mouton de Gruyter, pp. 765-85
Stirling, Lesley. 1988. Switch-Reference and Logophoricity in Discourse Representation Theory.
Edinburgh: University of Edinburgh Press.
Testelets, Yakov. 2008. Typology of Anaphoric Elements. Handout.
Thompson, Sandra A., Robert E. Longacre & Sin Ja Hwang. 2007. Adverbial clauses. In: Timothy
Shopen (ed.): Language typology and syntactic description. Volume II: complex constructions.
Cambridge: Cambridge University Press, pp. 237-300.
Treis, Yvonne. 2010. Purpose-encoding strategies in Kambaata. Afrika und Übersee 91: 1-38.
Treis, Yvonne 2012. Switch-reference and Omotic-Cushitic language contact in Southwest Ethiopia.
Journal of Language Contact 5.1: 80-116.
Valenzuela, Pilar. 2003. Transitivity in Shipibo-Konibo grammar: a typologically oriented study.
Ph.D. dissertation, University of Oregon.
Van Valin, Robert D. & Randy J. LaPolla. 1997. Syntax: structure, meaning, and function.
Cambridge, U.K. ; New York, NY: Cambridge University Press.
Vries, Lourens de. 2005. “Towards a typology of tail-head linkage in Papuan languages”. Studies in
Language 29.2:363-384.
Vuillermet, Marine. 2014. The multiple coreference in Ese Ejja subordinate clauses. In: Rik van Gijn,
Dejan Matić, Jeremy Hammond, Saskia van Putten and Ana Vilacy Galucio (eds.) Information
structure and reference tracking in complex sentences. John Benjamins, pp. 341-372.
Watanabe, Akira. 2000. ‘Feature copying and binding: evidence from complementizer agreement and
switch reference. Syntax 3(3): 159-181.
Wacke, O. 1931. Formenlehre der Ono-Sprache. Zeitschrift für Eingeborenen-Sprachen 21: 161-208.
Waltz, Nathan. 1976. Discourse functions of Guanano sentence and paragraph. In: R. Longacre & F.
Woods (eds.): Discourse grammar: studies in indigenous languages of Colombia, Panama, and
Ecuador, volume 1. Norman, OK: Summer Institute of Linguistics, pp. 21-145
Watkins, Laurel. 1976. The syntax and semantics of Kiowa switch-reference. In: R. Brown, K.
Houlihan, L. Hutchinson & A McLeish (eds.) Proceedings of the 1976 Mid-America Linguistics
Conference. Minneapolis: University of Minnesota, pp. 429-436.
Watkins, Laurel. 1993. The Discourse Function of Kiowa Switch-Reference". International Journal of
American Linguistics 59(2): 137-164
Weber, David J. 1980. Switch-reference: Quechua. In: Pamela Munro (ed.). Studies of switch-
reference. Los Angeles: University of California, pp. 48-64.
Weisser, Phillip. 2012. Is there switch-reference marking in coordinated clauses? In: P. Weisser (ed.)
Perspectives on switch-reference: local modeling and empirical distribution. Leipzig: Institut für
Linguistik, pp. 165-190.
Winter, W. 1976. Switch-reference in Yuman languages. In: M. Langdon & S. Silver (eds.) Hokan
studies. The Hague: Mouton, pp. 165-174.
Woodbury, Anthony C. 1983. Switch-reference, syntactic organization, and rhetorical structure in
Central Yup’ik Eskimo. In: John Haiman & Pamela Munro (eds.) Switch reference and Universal
Grammar. Amsterdam & Philadelphia: John Benjamins, pp. 291-316.
Recommended